Date : Signature of the Guide ... I thank Dr. J. Surya Prasad, Director/Principal, PESIT Bangalore South Campus for giving ...... 11th Annual International.
A STUDY ON QUALITY COMPLIANT CROSS LAYER ROUTING AND ALLOCATION STRATEGIES IN WIRELESS MESH NETWORKS
Submitted in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
by
SARASVATHI V
May, 2016
DECLARATION
I hereby declare that the thesis entitled “A STUDY ON QUALITY COMPLIANT CROSS LAYER ROUTING AND ALLOCATION STRATEGIES IN WIRELESS MESH NETWORKS” submitted by me, for the award of the degree of Doctor of Philosophy to VIT University is a record of bonafide work carried out by me under the supervision of Dr. N. Ch. Sriman Narayana Iyengar.
I further declare that the work reported in this thesis has not been submitted and will not be submitted, either in part or in full, for the award of any other degree or diploma in this institute or any other institute or university.
Place : Vellore Date :
Signature of the Candidate
CERTIFICATE
This is to certify that the thesis entitled “A STUDY ON QUALITY COMPLIANT CROSS LAYER ROUTING AND ALLOCATION STRATEGIES IN WIRELESS MESH NETWORKS” submitted by SARASVATHI V (School of Computer Science and Engineering) VIT University, for the award of the degree of Doctor of Philosophy, is a record of bonafide work carried out by her under my supervision, as per the VIT code of academic and research ethics.
The contents of this report have not been submitted and will not be submitted either in part or in full, for the award of any other degree or diploma in this institute or any other institute or university. The thesis fulfills the requirements and regulations of the University and in my opinion meets the necessary standards for submission.
Place : Vellore Date :
Signature of the Guide
ABSTRACT In recent years, the Multi-Radio Multi-Channel Wireless Mesh Network (MRMCWMN) is considered a reliable and cost effective way for Internet access in wide area. The key research challenge in MRMC-WMN is implementing an efficient channel assignment algorithm and routing techniques. The existing channel assignment algorithms limit the entire network to operate only in orthogonal channels (OC). Recent analysis and test-bed case studies in spectrum management proved that the partially overlapping channels (POC) with spatial reuse protocol integrated significantly improves the throughput and also eliminates the scarcity of the spectrum. The major pitfall with POC is its interference, hence finding a least interference path and assigning radio to it becomes a challenge. In this dissertation, the channel assignment is represented as a graph edge coloring problem using partially overlapping channels. Subsequently, a new routing metric called signal-to-noise plus interference ratio is proposed. Another major challenge in a MRMC-WMN is, finding a Quality of Service (QoS) satisfied and interference free path from the redundant paths, for transmitting the packets through that path. In this research work, an optimal intelligent routing using hybrid Particle Swarm Optimization-Genetic Algorithm (PSO-GA) is elaborated, which well meets the QoS constraints, and moreover it integrates the strength of PSO and GA. An intuitive Coefficient of Restitution based Cross layer interference aware Routing protocol (CoRCiaR) is introduced to improve TCP performance in Wireless Mesh Networks. The modified RTS/CTS algorithm and Round Trip Time (RTT) calculations ensure the reliability of WMN. Wireless Mesh Sensor nodes are deployed in harsh environments, like industrial Wireless Mesh Sensor Networks (IWMSN), where the equipment is exposed to temperature and electrical noise, so providing a reliable, interference free and efficient communication in this environment is a challenge. A Multi Route Rank based Routing (MR3) protocol is proposed, which enhances the link dynamics for IWMSN and also provides interference free reliable packet delivery in harsh environments.
i
ACKNOWLEDGEMENT First and Foremost, I thank God almighty for giving me good health, strength, patience and spirit to accomplish this thesis. I have been amazingly fortunate to work under the guidance of Dr. N. Ch. S. N. Iyengar, Senior Professor, SCSE, VIT University who has been an exemplary mentor and inspirator. Thank you very much sir for your expert guidance and encouragement throughout this research work. I am extremely grateful to Dr. Snehanshu Saha, Professor, Department of CSE, PESIT Bangalore South Campus for his scholarly inputs and inspiring ideas. I appreciate your support and insightful suggestions; I have learned many things from your research expertise. My sincere thanks to Dr. T. Arunkumar, Dean, School of Computer Science and Engineering for approving documents and completion of my research work. I thank Dr. Babu, Associate Dean, Academic research, VIT University for his support and encouragement. I would like to thank doctoral committee members Dr. S. N. Jagadeesha, Professor & HOD, Department of Computer Science & Engineering, J.N.N. College of Engineering, Dr. G. K. Patra, Scientist, CSIR, Centre for Mathematical Modelling and Computer Simulation, NAL Belur Campus and Dr. N. Jaishankar, SCSE, VIT University for their valuable inputs and comments. I would like to thank Dr. G. Viswanathan, Chancellor, VIT University, for facilitating the resources required for completing my research work. I also thank ViceChancellor and Vice-Presidents, VIT University, for their support. I thank Dr. J. Surya Prasad, Director/Principal, PESIT Bangalore South Campus for giving permission to pursue Ph.D in VIT University. I am very grateful to Dr. Srikanta Murthy K, Professor & Head, Department of Computer Science and
ii
Engineering, PESIT Bangalore South Campus for his constant support and motivation during my research work. I would like to extend my gratitude to all my colleagues of CSE Department, PESIT Bangalore South Campus, especially Mrs. Shubha Raj K. B and Mrs. Jermin Jeaunita T. C for their support and cheering me at every step of my research work. I would like to thank my source of spirit and supportive better half Mr.Ramasubramanian, with his confidence this thesis is possible. His patience for proof reading in every page of my paper and constant demand for excellence amazed me. This work would not have been possible without the support of sweet little heart, my son, Rohit Krishna for understanding my time constraints. Last but most importantly, I would like to thank my parents Mr. V. Varadharajan and V. Kanagavalli for their love, countless sacrifice and support in every part of my life. My special thanks to my sister Mrs. Menaka and my brother Mr. Venkatraman for their encouragement, support. I dedicate this research work to my family.
iii
Contents viii
List of Figures
Chapter 1
List of Tables
x
Introduction
1-15
1.1
Overview
1
1.2
Wireless Local Area Networks
2
1.2.1
2
Multi-Hop Wireless Networks
1.3
Wireless Mesh Networks
4
1.4
Types of WMNs
5
1.4.1
Flat Wireless Mesh Networks
5
1.4.2
Infrastructure Wireless Mesh Networks
5
1.4.3
Hybrid Wireless Mesh Networks
5
1.5
1.6
Challenges in WMN
6
1.5.1
Throughput Capacity
6
1.5.2
Reliability
7
1.5.3
Scalability
7
1.5.4
Resource Management
7
Cross Layer in WMN
8
1.6.1
Loosely Coupled Cross Layer Design
8
1.6.2
Tightly Coupled Cross Layer Design
8
1.6.3
PHY/MAC Cross Layer Design
9
1.6.4
Network/MAC Cross Layer Design
10
1.6.5
PHY/Transport Cross Layer Design
10
1.7
IEEE 802.11b/g Overview
10
1.8
Motivation
12
1.9
Organization of the Thesis
13
iv
1.10 Chapter 2
Chapter 4
16-27
Related work 2.1
Chapter 3
15
Summary
Literature on Channel Assignment
16
2.1.1
16
Centralized CA
2.1.1a Graph Based CA
17
2.1.1b Network Flow Based CA
17
2.1 1c Network Partitioning Based CA
18
2.1.2
18
Distributed CA
2.1.2a Gateway Centric Approach
18
2.1.2b Peer Oriented CA
18
2.2
Routing Metrics
20
2.3
Optimal Routing Algorithms
22
2.4
Congestion Control Algorithms
24
2.5
Summary
27
Centralized Rank Based Channel Assignment for MRMC WMNs
28-35
3.1
Introduction
28
3.2
Multi Radio Multi Channel Scenario
29
3.2.1
29
Challenges in MRMC WMN
3.3
System Architecture
30
3.4
Channel Assignment Algorithm
31
3.5
Simulation
33
3.6
Conclusion
35
An Efficient Interference Aware Partially Overlapping Channel Assignment and Routing
36-58
4.1
Introduction
36
4.2
Interference
38
4.2.1
38
Co-Channel Interference v
4.3
4.4
4.5 Chapter 5
Self-Interference
39
4.2.3
Adjacent Channel Interference
40
System Model
40
4.3.1
Channel Assignment Algorithm
41
4.3.2
Complexity of Edge Coloring
43
4.3.3
Routing Based on SINR Computation
46
4.3.4
Problem Formulation
48
Simulation
51
4.4.1
Simulation Results: Channel Assignment Algorithm
51
4.4.2
Simulation Results of Routing Using SINR
53 57
Conclusion
QoS Guaranteed Intelligent Routing Using Hybrid PSO-GA
59-75
5.1
Introduction
59
5.2
Routing Algorithms
61
5.3
Particle Swarm Optimization
61
5.4
Genetic Algorithm
63
5.5
Hybrid PSO-GA
65
5.6
The Mathematical Model for QoS Intelligent Routing
66
5.6.1
Fitness Functions
67
5.6.2
The Hybrid PSO-GA Routing
67
5.7
5.8 Chapter 6
4.2.2
Simulation
70
5.7.1
72
Simulation Results
75
Conclusion
Coefficient of Restitution based Cross Layer Interference Aware Routing Protocol 6.1
76-97 76
Introduction
vi
6.1.1 6.2
Chapter 8
76
System Model
78
6.2.1
Cross Layer Approach
78
6.2.2
Model and Motivation
79
6.2.3
Length
81
6.2.4
Modified RTS/CTS mechanism
81
6.2.5
Cumulative RTT
85
6.3
Routing Algorithm
87
6.4
Simulation
90
6.4.1
Evaluation Criteria
91
6.4.2
Performance Evaluation using COR
95
6.5 Chapter 7
Congestion in WMN
97
Conclusion
A Multi Route Rank Based Routing Protocol for Industrial Wireless Mesh Sensor Networks
98-112 98
7.1
Introduction
7.2
Reliable Routing Algorithms
100
7.3
System Model
102
7.3.1
Constructing Forwarding Node Set
104
7.3.2
Actual Forwarding Candidate Set
106
7.3.3
Cooperative Data Forwarding
107
7.4
Simulation Results
108
7.5
Conclusion
112
Conclusions and Future Enhancements
113
8.1
Contributions
113
8.2
Future Enhancements
114
Publications References
vii
List of Figures Figure
Description
Page No
1.1
Categories of multi-hop wireless networks
3
1.2
Hybrid WMNs
6
1.3
Available eleven partially overlapped channels in 802.11b frequency band
11
2.1
Classification of knowledge based CA algorithm
16
2.2
Frequency based CA
19
3.1
Multi-radio multi-channel
29
3.2
Architecture of multi-radio multi-channel WMN
30
3.3
The aggregate network throughput for 10 flows
33
3.4
Number of channel vs. delay
34
4.1
Co-channel interference
39
4.2
Self-interference
39
4.3
Adjacent channel interference
40
4.4
Block diagram that represents the interaction between channel assignment algorithm and the SINR based routing algorithm
41
4.5
Connection graph
44
4.6
Interference aware edge coloring for random topology
46
4.7
The proposed routing algorithm based on SINR
50
4.8
Random topology
52
4.9
Square topology
52
4.10
Throughput vs. number of radios
53
4.11
Aggregate network capacity vs. number of channels
53
4.12
Shortest path from user to the gateway node
54
4.13
Packet delivery ratio vs. number of channels
55
4.14
End-to-end delay vs. number of channels
56
viii
4.15
Routing overhead
56
5.1
Genetic algorithm
64
5.2
The hybrid PSO-GA routing algorithm
69
5.3
Random topology
71
5.4
Number of iteration vs. value of fitness
72
5.5
Number of nodes vs. convergence time
73
5.6
The packet delivery ratio vs. number of nodes
74
5.7
The average end-to-end delay vs. number of nodes
74
6.1
Cross layer design
78
6.2
Delay in queue
86
6.3
Routing in mesh architecture
89
6.4
Routing using RTT as a metric
90
6.5
RTT against number of hops
92
6.6
Delay vs. number of hops
93
6.7
Throughput vs. number of hops
93
6.8
RTT vs. number of nodes
94
6.9
Throughput vs. number of nodes
94
7.1
Architecture of multi route rank based routing protocol
102
7.2
Multi-radio multi-channel in IWMSN
103
7.3
Forwarding RREQ packets along reliable path
105
7.4
RREP packets forwarding
106
7.5
Density vs. packet delivery ratio
110
7.6
Density vs. end-to-end delay
110
7.7
Throughput vs. density
111
7.8
Number of control messages vs. density
111
ix
List of Tables Table
Description
Page No
1.1
Flavours of WLAN
1.2
Channel frequencies in IEEE 802.11b/g
12
3.1
Channel overlap
34
4.1
Link cost estimation based on SINR
51
4.2
Performance improvement of SINR metric over ETT and ETX
57
5.1
Path taken at each iteration and fitness in hybrid PSO-GA algorithm
73
5.2
Performance improvement of Hybrid PSO-GA over PSO and GA
75
6.1
Simulation settings for CoRCiaR algorithm
91
6.2
Performance improvement of CoRCiaR over semiTCP-AP and semiTCP
95
6.3
COR values
96
7.1
Forwarding path and the average rank of the node
107
7.2
Simulation parameters for MR3P
109
7.3
Performance improvement of MR3P over AODV-ETX and REPF
112
2
x
SYMBOLS AND NOTATIONS 2G
Second Generation
3G
Third Generation
ACK
Acknowledgement
ACO
Ant Colony Optimization
AFCS
Actual Forwarding Candidate Set
AODV
Ad-hoc On Demand Distance Vector Routing
AOMDV
Ad-hoc on Demand Multipath Distance Vector
BFS
Breadth First Search
BSA
Backward Smart Ants
BSCA
Balanced Static Channel Assignment
CA
Channel Assignment
CAEPO
Channel Assignment Exploiting Partially Overlapping Channels
CBR
Constant Bit Rate
CoR
Coefficient of Restitution
CoRCiaR
Coefficient of Restitution based Cross Layer interference aware Routing
CSMA/CA
Carrier Sense Multiple Access with Collision Avoidance
CSROR
Cross Layer Secure and Resource aware On-Demand Routing
DSL
Digital Subscriber Line
ETT
Expected Transmission Time
ETX
Expected Transmission Count
FNS
Forwarding Node Set
FSA
Forward Smart Ants
GA
Genetic Algorithm
GPRS
General Packet Radio Switch
GPS
Global Positioning System xi
HSA
Hello Smart Ants
IoT
Internet of Things
IWMSN
Industrial Wireless Mesh Sensor Networks.
JOCAC
Joint Optimal Channel Assignment and Congestion Control
LCI
Link Co-Channel Interference
M2M
Machine to Machine
MAC
Medium Access Control
MANETs
Mobile Ad-hoc Networks
MCG
Multi-radio Conflict Graph
MMAC
Multi-Channel MAC
MIMO
Multiple-Input Multiple-Output
MR3P
Multi Route Rank based Routing Protocol
MRMC
Multi-Radio Multi-Channel
NIC
Network Interface Card
OCARI
Optimization of Communication for Ad hoc Reliable Industrial network
OR
Opportunistic Routing
PCL
Preferable Channel List
PDA
Personal Digital Assistant
PDCA
Packing Dynamic Channel Assignment.
PDR
Packet Delivery Ratio
POC
Partially Overlapped Channel
PSO
Particle Swarm Optimization
PSO-GA
Particle Swarm Optimization-Genetic Algorithm
QoS
Quality of Service
Q-SMS
QoS-aware Shortest Multipath Source
REPF
Reliable and Efficient Packet Forwarding
xii
RREQ
Route Request
RREP
Route Reply
RERR
Route Error
RTS/CTS
Request to Send / Clear to Send
RTT
Round Trip Time
SA
Simulated Annealing
SINR
Signal-to-Interference plus Noise Ratio
TCP
Transmission Control Protocol
UDP
User Datagram Protocol
VoIP
Voice over IP
WCETT
Weighted Cumulative Expected Transmission Time
WLAN
Wireless Local Area Networks
WMNs
Wireless Mesh Networks
xiii
Chapter 1
Introduction 1.1 OVERVIEW Nowadays Internet access has become an essential part of our life and the mobile Internet access is very much required in every move of the life, starting from Global Positioning System (GPS) to transport schedules and bookings on the go. Social networking services such as LinkedIn, Google+, Facebook, Twitter, Instagram is a platform to share similar interests, ideas and real life events, and it provides an environment for a group communication among the friends. The World Wide Web is a source for all information about various fields such as science, medical, technology, politics, news, geographic details, agriculture, stock details and airline fares. The communication platform connects people together, facilitates talk to their family, relatives, and friends through Voice over IP (VOIP) (Mockapetris, 2006) calls and video chats, share their memorable moments through photo and video sharing. Consumers are using the Internet to buy electronics items, clothes, groceries, and other accessories online, and Internet plays a vital role in personal banking and utility payments. Companies display their product details in websites, and the order processing, bill payment and shipment tracking is moved from traditional way to e-commerce. Aircrafts provide inflight Internet access to their passengers, and the Internet is being used in all the facets of human life for various activities. The rapid development of mobile phones, gadgets, smart devices and Personal Digital Assistant (PDA) increases the need for research in wireless networks. The radical change in wireless communication is a game changer the way Internet is accessed, from the traditional wired network to anytime, anywhere. The wireless mesh network (WMN) (Akyildiz and Wang, 2005) is an emerging technology for broadband Internet access for cities, rural areas, and creating smart cities with limited or low infrastructure. Low power WMN trends as an attractive proposition for paradigm shifting technologies like IoT and M2M. Any wireless network forming a full or partial mesh topology is called WMN, and it is widely acknowledged as robust and cost-effective network infrastructure. 1
The WMNs are the next generation networks that seamlessly provide wireless connectivity to the organizations, institutes, enterprises and municipals. The diverse applications powered by WMNs are military field operations, metropolitan wireless access zones, public safety organizations and mining. There has been a multi-fold increase in the number of clients utilizing WMNs for Internet access. The robust fault tolerance, self-configuration and self-healing properties of the WMN provide a reliable communication.
1.2 WIRELESS LOCAL AREA NETWORKS (WLAN) Wireless LAN offers high flexibility to access the Internet within a building, school, library and office complex. The aim of WLAN is to replace the legacy wired line connection with the wire free flexible installation. Of course, there are some merits and demerits compared to the wired connection. The merits include: no planning required for installation, flexible device placement anywhere, and the robustness; and the demerits are: challenge in providing Quality of Service (QoS), interference with transmission, and WLAN can be hacked easily. The WLAN is included in IEEE 802.11 standard and the flavours of WLAN are indicated in Table 1.1. The recent trending research problems in IEEE 802.11 are: Radio resource management and channel allocation. Table 1.1. Flavours of WLAN
Protocol
Year of Release
Maximum data transfer speed
Frequency
Channel Bandwidth
802.11a
1999
54 Mbps
5 GHz
20 MHz
802.11b
1999
11 Mbps
2.4 GHz
20 MHz
802.11g
2003
54 Mbps
2.4 GHz
20 MHz
802.11n
2009
65 to 600 Mbps
2.4 & 5 GHz
20 & 40 MHz
802.11ac
2013
6.93 Gbps
5 GHz
20,40,80 & 160 MHz
1.2.1 MULTI-HOP WIRELESS NETWORKS The multi-hop wireless networks take two or more hops to transmit a message from sender to receiver, and the intermediate nodes are involved in forwarding the 2
packets. The advantages of multi-hop wireless networks are: cheaper installation cost due to less cable, cost effective, wide coverage area, operates in limited transmission power, and the robustness. The categories of multi-hop networks are indicated in Fig 1.1.
Fig 1.1. Categories of multi-hop wireless networks
The categories are: 1. The ad hoc wireless networks 2. WMNs 3. Wireless sensor networks 4. Hybrid wireless networks. The ad hoc networks are untethered and infrastructure less wireless networks. In this network, the collection of mobile nodes provide ad-hoc connectivity with dynamic topology and the individual nodes in the network act as a router, as well as an end host. The primary applications of ad hoc networks are: rescue operations, home networks, conferencing, and tactical communication. The WMNs are infrastructure based static networks with mesh topology. The Wireless Sensor Networks (WSNs) are a cluster of 3
mini sensor nodes with one or more base stations to collect the information. The hybrid wireless networks combine the traditional networks together to form a hybrid model, such as mobile networks with IP, infrastructure and peer-to-peer networks.
1.3 WIRELESS MESH NETWORKS (WMNs) Two types of nodes form the WMN: mesh routers and mesh clients. In mesh networking, mesh routers are used to carry out routing and forwarding functions, and these types of nodes are found to be static most of the time. The WMN primarily acts as a mesh backbone for mobile clients and it is in turn responsible for establishing the mesh connectivity among the clients. The gateway capabilities of the router integrate WMN with Wi-Fi, cellular, and other networks. The route selection is purely based on the link discovery and due to that fact, the link which does not interfere with other transmission would yield higher throughput. The mesh clients are simple nodes with one Network Interface Card (NIC), less mobility and routing capabilities. Every node in the network is involved in multi-hop communication and the routing process. In general, all the nodes in the network are connected wirelessly, except one or two gateways which are linked to the global network through wired connection. The characteristics of the WMNs are:
The network topology is moderately static.
Mobility of the node is low and so do the energy constraint.
Fixed infrastructure, only a little planning is required to deploy WMNs.
A central access point (AP) is not needed for communication.
The coverage is increased when more AP is installed, and the robustness of the link is also proportionally increased.
The low cost off-the-shelf multi radio nodes are used, as it is cost effective.
Access to heterogeneous networks is possible in WMNs. This feature is important for integrating WMN with future wireless networks.
Low transmission power, hence effective solution for harsh and dynamically changing environments are needed.
The nodes are equipped with self-healing, self-configuration capabilities.
4
1.4 TYPES OF WMNS There are three classes of network design architecture in WMN: flat, backbone and hybrid WMN. 1.4.1 FLAT WIRELESS MESH NETWORK It is a collection of client nodes and the nodes operate as end hosts and routers. The client nodes can either be a smart phone, laptop, PDA or any other mobile devices. In this setup, there is a peer to peer connectivity between client nodes. The client nodes synchronize themselves and offer array of functions, which include routing, configuration functions and application services to the customers. This type of network is very simple in nature and very similar to ad-hoc wireless networks. The primary disadvantage of this network is that, it is not scalable and having resource constraints. The major concern in the planning of flat WMN is its addressing strategy as it has significant impact on the scalability. 1.4.2 INFRASTRUCTURE WIRELESS MESH NETWORK The backbone or infrastructure is managed by wireless mesh routers, where the client nodes originate and terminate the data traffic. The client nodes are linked to the Internet through the mesh routers with gateway capability. This sort of infrastructure WMNs make up backhaul connectivity for conventional clients, conventional networks and other radio technologies. If the conventional client is not compatible with mesh routers, then the client would be diverted to any of the available access points and that in turn would have connected to the mesh routers. 1.4.3 HYBRID WIRELESS MESH NETWORK A hybrid wireless mesh network integrates the backbone and flat wireless networks as in Fig 1.2. The other infrastructure dependent networks such as mobile networks, satellite networks and Wi-Fi networks are combined with the backbone networks and, the connectivity between the nodes and Internet is achieved directly or through backbone networks. There are multiple technologies involved in designing backbone and hybrid WMN, and play a significant role in the evolution of WMN.
5
Fig 1.2. Hybrid WMNs (Akyildiz and Wang, 2005)
1.5 CHALLENGES IN WMN 1.5.1 THROUGHPUT CAPACITY The nodes in typical WMNs are mapped with single NIC that is assigned to a single channel. The throughput of a wireless network is limited by ( W ) (Gupta and n
Kumar, 2000), where n is the count of nodes in the network and the transmission at W bits per second. The attainable capacity of randomly placed nodes in a static wireless network is limited by ( (
W n1/ d
W n log n
) . The capacity of the arbitrary network is relative to
) , where d is the area of the network. Therefore, the throughput is low, when
there is an increase in the number of nodes in the wireless network. To support the huge traffic made by evolving applications and to enhance the capacity of WMNs, mesh node is armed with Multiple Radios (MR) and each interface is assigned with different frequencies. Due to inexpensive hardware cost (Bahl et al., 2004), the recent IEEE 802.11b/g standard nodes are built in with multiple radios.
6
The throughput decreases rapidly with increase in the number of hops in the network. The issues that adversely affect the throughput are: the design of Medium Access Control (MAC) protocol, hidden node and exposed node problems, unpredictable nature of the wireless channel and the error rate. The other factors that could potentially impact the throughput are the design of routing protocol and the greediness of the first node. All these factors are more severe in a single channel WMN in comparison to a multi-channel WMN. 1.5.2 RELIABILITY The wireless networks that operate in IEEE 802.11b/g standard uses the unlicensed electromagnetic spectrum, which interfere with devices operating in the same frequency band, i.e. the cordless phone and the microwave oven. The interference causes blocking of transmission signals and frequent disconnections; hence the channel error rate is higher, compared to the wired networks. The reliability and QoS are big challenging factors in wireless networks. But the mesh topology provides route diversity and reliability to deal with the unreliable communication. 1.5.3 SCALABILITY One of the very important requirements in WMNs is scalability (Huang and Lai, 2002) which attributes to the network performance and is evident with substantial increase in the nodes in a network. For the growth and evolution of any technology, scalability is considered as a key factor, and the protocols in all the layers co-operatively guarantee the scalability of the network. The primary scalability issues are: The Channel Assignment (CA) algorithms in MAC, the routing algorithm in network layer, and the reliable transmission in Transmission Control Protocol (TCP) layer. 1.5.4 RESOURCE MANAGEMENT The network resources are interfaces, channels and the energy. Resource management effectively utilizes these network resources and involves in the improvement of network characteristics like bandwidth, latency in WMN. The MAC protocol CSMA/CA (Carrier Sense Multiple Access with Collision Avoidance) is based on the shared channel access with back off timers, many flows in the network sharing the 7
same route and multiple connections exist in TCP layer. Load balancing across all the links, bandwidth stripping, and packet scheduling can help utilizing the resources properly. So, an efficient and optimal resource management algorithm is required across the protocol stack, from the physical layer to the application layer.
1.6 CROSS LAYER IN WMN The conventional layered module approach is not optimal for WMNs, thus the performance of network protocols can be increased by applying the cross layer technique. (Bhatia et al., 2004; Kozat et al., 2004; Chiang et al., 2004; Jiang et al., 2005) There are two types of cross layer design:
Tightly coupled.
Loosely coupled.
1.6.1 LOOSELY COUPLED CROSS LAYER DESIGN The parameters in physical and MAC layers are informed to the upper layers. So those lower layer parameters can be used to devise a performance improvement strategy for upper layers. For example, the loss rate and channel quality parameters of lower layers are used to enhance the performance at network and transport layers, by modifying the routing algorithm and TCP protocol (Holland et al., 1999; Liu et al., 2001). The cross layer approach can be accomplished between two or more layers. 1.6.2 TIGHTLY COUPLED CROSS LAYER DESIGN Passing a few network parameters between layers are not adequate in tightly coupled cross layer method. Algorithms in multiple layers are combined together to optimize the performance. For example, the routing of network layer and the channel allocation of MAC layer are combined in MRMC WMNs. Since the layers are combined, there is no need for exchange of parameters between the layers, hence the overhead is reduced and yields better performance than loosely coupled approach. In general, two to three layers are combined in a given protocol stack. The most complex case in this method is to combine different algorithms in a single layer, which is still an open research topic in WMNs. 8
The following are the most commonly used cross layer designs: 1. PHY/MAC cross layer design 2. Network/MAC cross layer design 3. PHY/Transport cross layer design. 1.6.3 PHY/MAC CROSS LAYER DESIGN The physical layer parameters such as interference, channel fading and noise ratio can significantly impact the network performance. The link adaptation technique is used to change the transmission rate and also to select the distinct modulation and coding patterns. The packet loss rate and the retransmission rate in MAC layer vary when the channel condition changes. The adaptive framing and rate adaptation techniques combinedly determine the best transmission rate and also help in selecting the suitable frame size for the transmission. The rate adaptive framing method is proposed by Chen et al. (2007), where the frame size is decided by the destination and passed to the source. The directional antenna in the node is used for long distance communication and it also attains very good link quality compared to omnidirectional. The directions of neighboring nodes are determined by the MAC layer and subsequently the PHY layer is engaged in tuning the antenna to a particular direction. The nodes collectively coordinate to determine its transmit power to utilize the benefits of directional antenna. The direction of adaptive antenna is significant to boost the network throughput with the cooperation of three layers such as network, MAC and physical layers. The Multiple-Input and Multiple-Output (MIMO) system significantly increases the link capacity (Zorzi et al., 2006). Its characteristics are:
Transmit diversity: Multiple antennas in a node send out the same data for reliability.
Spatial Multiplexing: All the antennas are utilized simultaneously to transmit various packets.
Beam forming: The transmission angle is used to focus on a definite direction to intensify the data rate.
Interference nulling: The interference is reduced by controlling different antennas. 9
1.6.4 NETWORK/MAC CROSS LAYER DESIGN The network and MAC layers are jointly optimized to find the routing path. The network layer collects traffic, interference, link characteristics from the MAC layer to discover the optimal route. The major functions of MAC layer are: allocating the network resources such as medium, time slots, channels and so on. The resource allocation is combined with the routing algorithm for efficient allocations. The QoS routing, described by (Chen et al., 1997; Lin and Liu, 1999), performs joint optimization between routing and code (time slot) allocation methods. The combined channel assignment and routing methodology deliver very good network performance, when multiple channels are used in the network. The Hyacinth protocol (Raniwala and Chiueh, 2005), combines the load balance routing with channel allocation, and it constructs a spanning tree for the route. The advanced features like rate control are integrated with routing for sophisticated resource allocation. 1.6.5 PHY/TRANSPORT CROSS LAYER DESIGN Due to the dynamic nature of link quality in WMN, the congestion control algorithms need to be cross refined with physical layer. The PHY layer parameters, such as transmit power, antenna direction, coding, etc., are collated and passed to the transport layer to regulate the end-to-end flow. The loss of packets due to congestion or bad link quality, caused by physical layer parameters are identified and altered, for instance, the transmission power to regulate the congestion and the coding rate to improve the quality of the link. Chiang (2005) described the joint power control with congestion control algorithms, which maximize the nodes utilization, subject to transmission rate and power of the source node. Coding, modulation and route are fixed in this algorithm; otherwise better optimal algorithm is required to adjust these parameters.
1.7 IEEE 802.11b/g OVERVIEW The IEEE 802.11b/g standard uses unlicensed spectrum that operates in 2.4 GHz band and the frequency channels 1 to 11 are used for data transmission, of which the channels 1, 6 and 11 are called orthogonal channels (OC) (Fig 1.3). The end-host and 10
access point antennas within the range are organized into carrier frequency of the channel to use a specific channel. The channels 1 and 2 operate in frequency bands 2.412 GHz and 2.417 GHz respectively. The carrier frequency variance between the channels is about 5 MHz, but the dominance of the signal in the frequency spectrum is approximately 22 MHz. This signifies that when the channels 1 and 2 used for data transmission by the adjacent channels, there would be overlaps and that results in severe interference and data drop.
Fig 1.3. Available eleven partially overlapped channels in 802.11b frequency band
There can only 1 out of 5 consecutive channels be used simultaneously on the available bandwidth in multi-channel scenario. This indicates that only 3 non-overlapping channels are available to avoid the interference with the current utilization techniques. Today, hundreds of devices such as microwave ovens, cordless, etc. function in the same frequency around the end devices that leads to major issues like frequent interference and density. The solution defined for this problem is the partially overlapped channel (POC) assignment (Franklin et al., 2011; Cui et al., 2011; Wang et al., 2011), where the channels can be chosen with 2 other channels apart, and two simultaneous transmissions are 10 m apart. This solution also delivers the same throughput in analysis (Mishra et al., 2006) in comparison to the existing non-overlapping channel assignment technique. When it comes to densely populated wireless mesh network, all the available channels that are assigned with spatial separation can be used to gain higher performance than the orthogonal channels. 11
Table 1.2. Channel frequencies in IEEE 802.11b/g Channel
Center
Minimum
Maximum
Frequency(MHz)
Frequency(MHz)
Frequency(MHz)
2412 2417 2422 2427 2432 2437 2442 2447 2452 2457 2462
2401 2405 2411 2416 2421 2426 2431 2436 2441 2446 2451
2423 2428 2433 2438 2443 2448 2453 2458 2463 2468 2473
1 2 3 4 5 6 7 8 9 10 11
1.8 MOTIVATION The Multi-Radio Multi-Channel WMN (MRMC-WMN) offers a significant increase in the network capacity (Bahl et al., 2004; Zhang et al., 2005), but it faces many challenges and concerns. By considering the wide range of applications provided by WMN, an innovative research study in MRMC-WMN is presented in this thesis. The primary focus of this research is to enhance the throughput, scalability and reliability in WMN. The nodes in the WMN are relatively static, so there are no restrictions on power consumption. A protocol designed for an ad hoc setup is less suitable for WMNs, because the nodes in the ad hoc networks have high mobility, but on the other hand, the nodes in WMNs are with less mobility. The objectives of this thesis are:
The scarceness of electromagnetic spectrum decreases the capacity of WMN. In order to increase the number of users accessing the Internet, the full range of frequency is used with spatial reuse. So, an efficient CA algorithm is required to use all the available channels.
Finding a route with guaranteed QoS, and to find an interference free path from the redundant paths, for efficient transmission of the packets through the identified path.
Providing more reliability in WMN and to boost the TCP performance in WMN. 12
The primary focus areas of this research work are: the channel allocation strategies, finding a new routing metric and routing algorithm for MRMC wireless networks which promises high performance and QoS guarantee in any densely deployed networks.
1.9 ORGANIZATION OF THE THESIS The thesis entitled “A Study on Quality Compliant Cross Layer Routing and Allocation Strategies in Wireless Mesh Networks” is organized into eight chapters. Chapter 2 presents the research proposal in centralized and distributed CA algorithms; Existing routing and link metrics in WMN are discussed. Also, meta-heuristic algorithms for optimal routing, and the existing congestion control algorithms are elaborated. A centralized rank based CA algorithm is proposed in Chapter 3. This algorithm assigns the load aware metric ‘Weighted Cumulative Expected Transmission Time’ (WCETT) as a link cost in DSR and AODV routing algorithms to enhance the throughput. Every node is assigned with a rank which is estimated by the following three parameters such as load on a link, WCETT from node to the gateway, and the number of interfaces in a node. In recent years, MRMC-WMN is considered a reliable and cost effective way for Internet access in wide area. The key research challenge in MRMC-WMN is implementing an efficient channel assignment algorithm and routing techniques. The existing channel assignment algorithms limit the entire network to operate only in OC. Various channel assignment algorithms and strategies have been analyzed and experimented to enhance the throughput using OCs. Recent analysis and test-bed case studies in spectrum management proved that the partially overlapping channels with spatial reuse protocol integrated significantly improves the throughput and also eliminates the scarcity of the spectrum. The major pitfall with POC is its interference, hence finding a least interference path and assigning radio to it becomes a challenge. In chapter 4, the channel assignment is represented as a graph edge coloring problem using
13
POC. Subsequently, a new routing metric called signal-to-noise plus interference ratio is proposed. The Particle Swarm Optimization (PSO) is an optimization technique to find candidate solutions in the search space optimally and it applies artificial intelligence to solve the routing problem. On the other hand, the Genetic Algorithm (GA) is a population based meta-heuristic optimization algorithm inspired by the natural evolution, such as selection, mutation and crossover. The PSO can easily fall into a local optimal solution, but at the same time GA is not suitable for dynamic data due to the underlying dynamic network. In Chapter 5, an optimal intelligent routing is proposed; it uses a hybrid PSOGA, which also meets the QoS requirements. Moreover, it consolidates the strengths of PSO and GA. The proposed Coefficient of Restitution based Cross layer Interference aware Routing protocol (CoRCiaR) in Chapter 6, improves the TCP performance in Wireless Mesh Networks. This approach comprises of two-steps: Initially, the interference detection algorithm is developed at MAC layer by enhancing the RTS/CTS (Request to Send / Clear to Send) method. The congestion is identified by Round Trip Time (RTT) measurements, and subsequently the route discovery module selects the alternative path to send the data packets. The packets are transmitted in the congestion free path seamlessly by the source. The performance of the CoRCiaR protocol is measured by the Coefficient of Restitution (COR) parameter. Wireless mesh sensor nodes are deployed in harsh environments, like Industrial Wireless Mesh Sensor Networks (IWMSN), where the equipment is exposed to temperature and electrical noise. So providing a reliable, interference free and efficient communication in this environment is a challenge. The proposed Multi Route Rank based Routing (MR3) protocol in Chapter 7, enhances the link dynamics for IWMSN and also provides an interference free reliable packet delivery in harsh environments. The rank of a node is estimated based on density, hop count, energy and signal to interference plus noise ratio (SINR). The route discovery phase finds the rank value to forward the data packet in a reliable path.
14
The conclusion of the thesis and scope is discussed in Chapter 8 along with the summary of investigation and future plan.
1.10 SUMMARY The need and the usages of Internet in various fields are described in the Introduction chapter. The basic concepts such as, the types of wireless local area networks, bandwidth of each network type, maximum transfer speed, and the four types of multi hop wireless networks are explained. The introduction of WMN, types, characteristics, challenges, merits and demerits of each type are presented. The merits of cross layer design and the significance of the design between various layers are analyzed. The available frequency band in IEEE 802.11b/g standard and the necessity for partially overlapping is demonstrated. Finally, the motivation of the research and the research problem is presented.
15
Chapter 2
Related Work There are plenty of channel assignments problems have already been proposed to boost the network throughput by diminishing the interference. But, most of the CA algorithms proposed are combined with either routing or congestion control techniques.
2.1 LITERATURE ON CHANNEL ASSIGNMENT In general, the CA is classified (Fig 2.1) into centralized and distributed methods, the algorithm is further divided into static, dynamic and hybrid methods. The static and dynamic CA methods mainly depend on the time interval between the successive runs of the protocol. In static CA, the node’s radio frequency is tuned for a certain channel permanently; though it is easy to implement and deploy, not that suitable for dynamic environments. In dynamic CA strategy, the interface can be assigned with different channels, and the frequency band shifts from one frequency to another for every data transmission.
Channel Assignment
Centralized
Graph based
Network flows
Distributed
Network partitioning
Gateway
Peer
Fig 2.1. Classification of knowledge based CA algorithm
2.1.1 CENTRALIZED CA In the centralized method, all the nodes in the network communicate about its current CA and the list of available channels ready to be assigned, to the gateway node. The gateway collects the data and retains the knowledge base about interferences for 16
future CA. The gateway is solely responsible for channel assignment to all the nodes in the network. The nodes in the network are modeled as a graph in the graph based approach. 2.1.1a GRAPH BASED CA The conflict graph model is used by Jain et al. (2003) to design the interference in in MRMC WMN. Marina and Das (2005) proposed a connected low interference CA and it greedily selects a channel using the edge coloring method, which uses only 3 orthogonal channels. The proposal made by Tang et al. (2005), INSTC (minimum INterference Survivable Topology Control) and Link Co-Channel Interference (LCI) metrics are defined, where the link is modeled as LCI metric. In another work, a breadth first search method for CA (BFS-CA) (Ramachandran at al., 2006) is introduced, where the channel selection is greedy on MCG (Multi-Radio Conflict Graph). Subramanian et al. (2008) described a CA using the conflict graph to reduce the co-channel interference and also uses vertex coloring method and Max-K cut problem. The merits of graph based method are intuitive and simple to design, but it does not consider the traffic into account. 2.1.1b NETWORK FLOW BASED CA The traffic information such as end-to-end traffic, load on the link are passed on to the central node by the individual nodes in the network. A Load Aware CA algorithm is proposed by Raniwala et al. (2004) which maps channel onto a radio by considering the proportion of the available bandwidth and the predictable traffic. The two sub problems in this algorithm are: The neighbor to interface mapping, an approach to identify a NIC on the node to communicate with its neighbors; and the interface to channel mapping approach, to determine the operational frequency of the NIC. Kodialam and Nandagopal (2005) jointly proposed BSAC and PDCA algorithms, which assigns channels to interfaces in such a manner that end-to-end traffic vector is achieved. Initially, the BSCA performs the static assignment and then applies the coloring method to allocate time slot to each channel. The PDCA performs the dynamic CA that switches the channel for every time slot. The LP framework is used in these algorithms to include 17
many objective functions. Alicherry et al. (2006) described a Joint Routing and CA, which models the interference and flow restrictions using a constant factor approximation method. The flow transformation method guarantees the maximum data that can be transmitted on a given route and also reduces congestion near the gateway. The above three studies carried out on a constant traffic rate, but it may not be able to handle the bursty traffic situation arises in the dynamic network environment. 2.1.1c NETWORK PARTITIONING BASED CA Brzezinski et al. (2006) presented the network partitioning approach which avails the benefits of MR capability for throughput maximization. The network is divided into small subnets, and then distributed scheduling method is applied. The local pooling method is used for partitioning to produce non-interfering subnets. 2.1.2 DISTRIBUTED CA 2.1.2a GATEWAY CENTRIC APPROACH The exchange of information and the cooperation among all the nodes in the network is essential in distributed CA algorithms. The routing metric is one of the key factors for providing information and decision making in CA. Most of the traffic is near the gateway as the nodes in WMN are connected to the Internet through the gateway. Raniwala and Chiueh (2005) proposed distributed CA and routing algorithms or Hyacinth, which uses the local flow information to bind the frequencies dynamically and routes the message. Many spanning trees are constructed to divert the traffic and to handle the node failure. Hyacinth architecture addresses the bandwidth problem by using the orthogonal channels. Gálvez et al. (2008) presented a distributed load balancing protocol which addresses many practical scenarios. As the gateways are coordinated themselves to balance the load, it is suitable for both stable and skewed topologies. This approach reduces the overhead of the gateway through effective load balancing in the networks. 2.1.2b PEER ORIENTED CA Peer oriented CA does not make any assumption about the traffic pattern, and can detect peer to peer load and any change in the network topology. Li and Xu (2009) 18
described the joint CA and routing for real time traffics in WMN. Each node in the network maintains the channel data such as the number of neighbors, the number of interfaces and its channel assignation, available channels and the channel bandwidth. During the node initialization process, the interface and its assigned frequency information are collected. The frequency information is embedded in RREQ packet itself and the packet is flooded in all the NICs in the node, in order to assign different channels to the adjacent interfaces, to avoid self-interference. The link layer protocol and a new metric, Multi-Channel Routing (MCR), are formulated by Kyasanur and Vaidya (2006) for handling MCs, by dividing the interfaces into fixed interface and switchable interface. The “fixed interfaces” are assigned to “fixed channels” for a long duration of time and the “switchable interfaces” can be assigned with different channels and also it is expected to be changed more often for every packet transmissions. The complexity of channel switching is concealed from the higher layers by the MAC layer. The ETT metric and the switching cost are merged in MCR metric. Rad and Wong presented (2006) a joint optimal CA and congestion control (JOCAC) algorithm for WMN. JOCAC uses OC as well as POC for the utility maximization problem. The congestion price of the link, transmission power, number of NICs and the available channels are taken into account in designing the JOCAC problem.
Channel Assignment
Static Assignment
Dymanic Assignment
Hybrid Assignment
Fig 2.2. Frequency based CA
Generally the proposed CA is classified into static, dynamic and hybrid methods (Fig 2.2). The static CA constantly assigns a channel to the NIC in the node, and CA is 19
carried out only once to select the channel. The Dynamic CA algorithm is re-executed in certain intervals to reassign the channel to NIC in the node. In hybrid method, some interfaces are assigned with static channel; and other interfaces are mapped to dynamic scheme.
2.2. ROUTING METRICS The interference in the link is observed and alleviated using efficient routing techniques and routing metrics. The traditional hop count metric can be simple to use and implement in routing algorithms, but the routing path built using the hop count cannot produce better solution in wireless networks, because the links are treated uniformly and the interference on the wireless link is discounted. The Expected Transmission Count (ETX) (Couto et al., 2003) metric integrates the interference and loss ratio to find the optimal route with high throughput for effective data delivery. The forward and backward delivery ratios are used to compute the link metric. ETX
1 df db
(2.1)
Where df and db are the forward and backward delivery ratios. The d f is the probability of packet correctly reaching at the destination. The db is the probability that the acknowledgement of the message is arrived successfully. The bandwidth, traffic and interference are not considered in the calculation; hence this metric can only be used in a single channel WMN, and it does not fit for MC WMN. Expected Transmission Time (ETT) metric finds the optimal route as it combines the bandwidth and packet size together with ETX. The ETT is also known as “bandwidth adjusted ETX” (Draves et al., 2004), and ETT of the link is computed by equation 2.2, where packet size is represented by P and the bandwidth is indicated by B. ETT
ETX P B
20
(2.2)
The ETT is derived without considering the traffic originating from the neighbor nodes, so the bandwidth is curtailed by the contending traffic and that results in packet loss and collision in the network. The weighted cumulative expected transmission time (WCETT) (Draves et al., 2004) is modeled to enhance the ETT metric. The hops that are assigned with different channels cause less interference compared to the hops assigned with same channel, so an additional term is included for interference along with the sum of ETT in WCETT metric. The channel diversity is an important factor in MRMC WMN; So WCETT is suitable for multi-channel scenario. The WCETT metric is computed by the equation 2.3 for a path p WCETT p (1 ) ETT j max x i jp
1i n
(2.3)
Where Xi is the sum of ETT of the links which are on channel i. n is the number of OC and the tunable parameter α lies between 0 and 1. The WCETT improves the performance of MRMC WMN, in comparison to the other three metrics discussed earlier. The demerit of WCETT is that the interference arises from the nodes which are competing for the same frequency is not considered in the calculation. Cherif et al. (2013) proposed LO-PPAOMDV routing algorithm, which incorporates Link Quality, MAC Overhead to differentiate the link loss emerges from congestion or mobility. This routing algorithm finds a new path before the original path cease to exist by monitoring the received signal strength. The channel capacity is subject to fading, so, Hasen et al. (2015) presented a closed form expression to improve the channel capacity, and that produces increase in the SNR level. The metrics developed for traditional networks are not suitable for MRMC WMN, as those metrics are not considering the interference as evaluation criteria. Each metrics have its own merits and demerits, and yields optimal route in a particular scenario. In this thesis, the CA problem is combined with routing and is solved using the graph edge coloring, and the SINR metric is proposed to sense the interference in each link and finally ensures that the throughput of individual link is maximized.
21
2.3 OPTIMAL ROUTING ALGORITHMS There are a quiet large number of meta-heuristics methods available for optimization of the routing algorithm. They fall into different categories as follows: Simulated Annealing, Evolutionary Algorithms, Tabu Search and Swarm intelligence Algorithms. Simulated Annealing is a meta-heuristic method with probability feature to find a good optimized solution in a large search space. It is an adaptable method for networks and more suitable for the discrete type; since nodes in the network are discrete, simulated annealing method can be applied here. The simulated annealing algorithm is more appropriate for time-bound solutions, it may not guarantee for the best solution, but it gives a better solution within a given time limit. Zhao et al. (2010) considered link instability and link quality for routing algorithm in WMN, and Ant Colony Optimization (ACO) and Simulated Annealing (SA) algorithms are presented. SA is integrated with the ACO to speed up the convergence rate. The improved Dijkstra’s algorithm is used in (Zuo et al., 2013) to find a shortest path between the end node and the gateway, which improved the reliability of data transmission and also dealt with link failures effectively. Here, the ACO is combined with enhanced Dijkstra’s algorithm, the route exploration and maintenance is performed by ACO and the route setup is carried out by enhanced Dijkstra’s algorithm. Initially, the network manager identifies the shortest route to each node by the enhanced Dijkstra’s algorithm. After that, the ACO algorithm explores the remaining routes in the route exploration stage, which is explored instantaneously along with the data transmission. The topological changes in the network are considered in the route maintenance scheme and the response is efficient in a timely manner. There are two types of pheromones used in this algorithm: The regular pheromone and the virtual pheromone. The goodness of the route is estimated by the regular pheromone and it also helps in finding the route which can be used for packet transmission. A virtual pheromone is presented to send the data which helps ants for sampling the possible paths in the network.
22
Bokhari et al. (2012) presented a distributed routing method for providing the high throughput, less interference and load balanced path in WMN. This AntMesh proposal is based on the stochastic approach, which simultaneously performs routing and data forwarding in a dynamic network. Ants utilize multiple channels efficiently with multiple interfaces in WMN. There are three types of ants used in this algorithm: Forward Smart Ants (FSA), Backward Smart Ants (BSA) and Hello Smart Ants (HSA). FSA travels from the source to the destination to determine the paths, BSA travels from the destination to the source for updating the routing tables and HSA collects the local link quality information to populate the link estimation table. Dia et al. (2008) have considered the dynamic nature of load or traffic for optimal routing. It is based on two of the components in the framework: traffic assessment and routing optimization. The future traffic demand is predicted from historical data by the time series analysis in the traffic assessment phase. The traffic estimation predicts the mean demand at a long period of time in dynamic networks and also statistical distribution. Two of the routing algorithms are presented in two different forms of traffic demand estimation. First, the mean value of traffic forecast is specified as input to the routing algorithm which is expressed as a linear programming problem to enhance the data flow. Second, the traffic demand is characterized based on a random variable to incorporate statistical distribution in problem formation. The routing optimization balances the traffic and ensures that minimum congestion will be incurred. Pries et al. (2010) focused on routing and CA in WMN to achieve the max-min fair throughput allocation. They have used genetic approach to optimize the deployment of WMN. The progressive filling algorithm is used to achieve the max-min fairness; eight different fitness functions are used for optimizing the route from the source to the gateway, such as minimum mean and maximum throughput. Instead of 2-Point Crossover, two new cross variants called Cell and the Subtree Crossover are introduced to produce the best solution. When a large number of users per gateway are used, the Sub tree Crossover method was chosen for better performance and the Cell Crossover given to the best solutions in scenarios with a less number of end-users per gateway.
23
In most of the work mentioned above, the authors focused on routing in WMN with orthogonal channels assigned to the interfaces, and only the physical distance between the source and the gateway is considered without considering the interference on the link. The capacity of the link is not only based on the physical distance, but also interference on the link. In contrast to the existing work, this thesis focuses more on routing in MRMC-WMN with POC assigned; radios are assigned with channels in such a way that it does not interfere with neighbouring links. Adjacent channel and self-interference is observed more when the partially overlapping channel is assigned to a radio. The shortest path may not be the interference free path, and it may lead to a collision domain, so, the hybrid PSO-GA algorithm is proposed in this thesis which finds less interference path efficiently and also satisfies the bandwidth, delay and jitter requests specified by the user.
2.4 CONGESTION CONTROL ALGORITHMS Basic TCP congestion control does not perform well in wireless networks because of the fact that difficult to differentiate between congestion and bit error. An improved TCP congestion control algorithm (Khurshid et al., 2011) for wireless networks is proposed. The basic TCP congestion control algorithm is modified to enhance the performance of TCP in wireless networks (Balakrishnan et al., 1999; Chandran at al., 2001). The multiplicative decrease is refined in TCP NewReno and the statistics counter is used to monitor the frequencies of timeout occurrences and 3 duplicate acknowledgements. The value of the counter and the quantum of time between two consecutive timeouts decide the congestion losses or bit error. This algorithm gives better performance in heterogeneous networks and the modification has been done only on the sender side of TCP, no burden on the internal network. If there is a real congestion, then it performs as original TCP NewReno, otherwise it carries out transmitting at a good speed. So, the capacity of the network is utilized properly in bit errors scenario. The packet arrival and departure time are compared, to distinguish between the congestion loss and error losses in Wireless TCP (Sinha et al., 1999). This is an end-to24
end semantic mechanism used in Wireless TCP (WTCP), and it does not half down its transmission rate like TCP, instead sending rate is adjusted at the receiver based on inter -packet delay metric. The WTCP uses the rate based transmission and the feedback is taken from the receiver to retransmit the packet. The TCP Westwood (Mascolo et al., 2001) upgrades the TCP Reno for wireless networks and it primarily depends on end-to-end bandwidth approximation to figure out the origins of packet losses. The inspection or interception of packets at proxy node is not required in TCP Westwood as it continuously monitors the ACK returning rate. The network capacity is calculated by measuring the arrival rate of ACK and the same is denoted by SBW[ j] .Also, the smoothed value BWE[ j] is calculated by low-pass filtering the sequence of SBW[ j] . SBW[ j]
BWE[ j]
Packet Size current _ time prev _ ACK _ time
(1 t ) * (SBW[ j] SBW[ j 1] t * BWE[ j 1] 2
(2.4)
(2.5)
Where t is the low-pass filter factor, packet_size specifies packet’s size, current_time is the most recent time, and prev_ACK_time is the time when ACK received. This method tries to estimate the approximate bandwidth to set the congestion window size. The TCP Vega (Brakmo and Petersen, 1995) uses the modified slow start mechanism and the new timeout mechanism for congestion avoidance. The objective of TCP Vega is to maintain the correct amount of data in the network. Based on the variation of estimated extra data present in the network, the algorithm decides the sending rate. If the source is transmitting too much of data, there will be a delay in getting the acknowledgement and it will lead to congestion. The TCP Vega finds the BaseRTT when the network is not congested, and in this case, the expected rate is given by
25
Expected Rate
Congestion Window BaseRTT
(2.6)
Where, the congestion window indicates the number of bytes in transition. The current sending rate is calculated by actual RTT. ActualRate
Congestion Window ActualRTT
(2.7)
The difference between the ActualRate and ExpectedRate is calculated and accordingly the congestion window is adjusted.
Diff ExpectedRa te ActualRate
(2.8)
The thresholds α and β are used to measure the amount of data present in the network. If Diff < α, then the congestion window is increased linearly. When Diff > β, the congestion window is decreased linearly. If α < Diff < β, then the congestion window is unchanged. However, the congestion control algorithms (Ammar et al., 2011; ElRakabawy and Lindemann, 2011; Jacobson, 1988) may not be appropriate for MRMC WMN with partially overlapping channels, where the packet loss is due to the interference and its dynamic nature of CA. Therefore, instead of the typical congestion control algorithm, CoRCiaR protocol is proposed in this thesis, which involves MAC and routing layers for reliable TCP protocol. The XCHARM (Chowdhury et al., 2013) cross layer routing protocol that chooses the transmission rate by combining the interference, and channel fading. It proposes the inter-channel model that determines the adjacent channel interference; the channel selection and the fading calculation are integrated into the routing protocol. The route is selected based on the channel which gives high data rates and less interference level. The latency of the path is estimated by packet error, contention, forward error correcting codes and the data rate on the selected channels. The route maintenance is proposed to monitor the network performance and trigger the recovery process in case of link failure. LO-PPAOMDV (Cherif et al., 2013) uses cross layer approach to find a congestion free route, by collecting information from MAC. The MAC informs 26
unsuccessful communication to the routing layer to identify the congestion using Lagrange interpolation. Chellaih et al. (2012) considered routing as a multi constraint problem and route is chosen based on more than one constraint such as buffer occupancy, energy and hop count.
2.5 SUMMARY The classification of CA algorithms, advantages and disadvantages of each method is described in this chapter. The cross layer between CA and routing, and the different routing metrics such as hop count, ETX, ETT, WCETT are described. From the existing CA algorithm and the routing metric, it is observed that the POC needs efficient CA algorithm and the routing metrics to avoid the interference. Some of the metaheuristic algorithm such as simulated annealing, ACO and genetic algorithms, and their merits and demerits are analyzed. The performance of TCP in wireless networks is poor, and the various congestion control algorithms such as wireless TCP, TCP Westwood, TCP Vega and TCP NewReno are carefully analyzed.
27
Chapter 3
Centralized Rank Based Channel Assignment for MRMC WMNs 3.1 INTRODUCTION The WMN is a distinct type of multi-hop network that preserves mesh topology among the nodes. The routers and the end-users (clients) collectively form the WMN, where the clients are ad-hoc in nature, but the routers are static and the provider of backbone connectivity. The router’s role is significant in WMN as it implements typical routing algorithms and establishes the mesh networking with extended routing functions. The automatic configuration feature of WMN brings in assorted benefits such as low maintenance, easy installation and fault tolerance. In WMN, the routers are interconnected themselves, and the backhaul access is achieved through a border router or the gateway. The MRMC in WMN provides great advantages in terms of network performance and the capacity than a single channel single radio. The router of the MRMC WMN is armed with multiple NICs, and each interface is mapped to different frequency. Numerous research proposals have been developed for CA, which typically fall into two categories. The first one, in which the NIC switches the channel frequently for every packet (Maheswari et al., 2006) and in the second approach, channels are permanently assigned to the NIC (Rad and Wong, 2007). Raniwala and Chiueh (2005) designed Hyacinth protocol which can be adapted with little modifications in the software to suit for 802.11 NICs. In Hyacinth protocol, the spanning tree is constructed to stabilize the traffic and also to divert the traffic in case of link break down. Das et al. (2005) proposed a static CA that focuses linear programming model for MRMC WMNs and it is not suitable for real traffic. Tang et al. (2006) developed a protocol for link scheduling and it delivers the maximum throughout; it also minimizes the noise level by increasing the signal strength. In this chapter, the rank is estimated for CA by using the load on the link, distance to reach the gateway and the number of interfaces in a node.
28
3.2 MULTI RADIO MULTI CHANNEL SCENARIO
Channel 1
Channel 1 Channel 6
Channel 6
Fig 3.1. Multi-radio multi-channel
In Fig 3.1, the node A and B are within the communication range of each other, so direct transmission and receiving is possible between the nodes. The nodes A and B are equipped with two NICs each and one NIC can be used for transmission and other one for reception. In addition to that if the radios are assigned with non-interference channel, the throughput is doubled. So, the channel assignment is one of the key factors for better performance. For example, the nodes A and B are assigned with non-overlapping channels 1 and 6, then better throughput is achieved, and on the other hand if both NICs are assigned with the same channel or adjacent channels, then the performance is compromised. 3.2.1 CHALLENGES IN MRMC WMN
The transmitting and receiving NICs in a node should have proper channel separation to avoid self-interference (Dhananjay et al., 2009; Rabinson et al., 2005).
The efficient channel assignment algorithm is required in MRMC WMN environment to pair up the channel and radio interfaces in a feasible manner.
The current MAC protocol is not compatible with MRMC WMN, so the MAC protocol is modified to support MRMC (Crichigno et al., 2007).
The routing protocol should be enhanced to take the full advantages of MRMC (Ramachandran et al., 2006; Wu et al., 2006). 29
The routing protocol should have knowledge of the list of channels involved in shortest path that lead to interference. So the interference in the link should also be considered as one of the routing metric.
3.3 SYSTEM ARCHITECTURE Fig 3.2 depicts the MRMC WMN system architecture that composed of routers, aggregation nodes and the gateway, where the routers and aggregation nodes form mesh connectivity among themselves. The routers perform basic routing and also send and receive data to end users and the global network through the gateway. The end users connect to access point (aggregation node) within its range.
Fig 3.2. Architecture of multi-radio multi-channel WMN
The aggregation nodes are responsible for accumulating the traffic from end users and pass those data to the Internet through the gateway. So, aggregation nodes estimate the traffic in the network that can be used for finding the rank. Only one or two gateways are directly connected to the Internet using a wired connection. The gateway estimates the traffic between each pair of nodes and also accumulates the traffic information. The routers are assigned with two NICs, so it can use maximum of two channels at a time. In this system architecture, 5 channels are considered. The end user is equipped with only
30
one NIC, the communication between the end user and the access point is through WiFi access.
3.4 CHANNEL ASSIGNMENT ALGORITHM If two nodes are assigned with the same frequency and happened to be within the communication range of each other, then direct connection is possible. The rank is calculated for all the nodes in the network based on WCETT, number of NICs and the load on a node. The CA algorithm primarily focuses MRMC-WMN with one gateway. With minor changes it can be used for multiple gateways. The objective of this algorithm is to increase the network throughput and preserve the connection between the end user and the gateway. The rank of the node is used to decide the frequency for the interface and the following parameters are considered for the rank calculation: • The total load of a node is calculated which is described by Raniwala et al. (2004). • WCETT from the gateway. • The number of interfaces on a node. Higher the rank indicates more traffic in the node. The rank is calculated using the equation 3.1 Rank=
Total Traffic in the node WCETT from the node to the gateway * Number of NICs in the node
(3.1)
The total load on the node greatly influences CA algorithm as it is directly proportional to the rank and also increases the rank. The node which carries less traffic is assigned to interfering channel and heavy traffic nodes are assigned with least interference channel. The most of the traffic is around the gateway, so the nodes closer to the gateway are most likely to be assigned with higher ranks. The less number of interfaces in a node fetches the higher rank. Due to dynamic traffic behavior, the rank is recalculated and the channels are reassigned every time. After the rank calculation, the algorithm traverses through the nodes in the order of the rank.
31
Algorithm Input: Connected graph Available Channels Number of Interfaces at each node Step 1.
Traverse each node u based on the rank value. Higher rank nodes are visited earlier.
Step 2.
For each link (u, v) If interfaces are not yet mapped with channels If both node u and v are assigned with the common channel C Map the common channel C to link (u, v)
Step 3.
While node u has an unassigned interface
Step 4.
Identify a neighbor node v which has more traffic
Step 5.
If node u has an unassigned interface
Step 6.
If node v has all its interfaces assigned Select the least used frequency (channel) C among those mapped to interfaces at node v Map C to an interface at node v Map C to the link (u, v)
Step 7.
Else if node v still has an unassigned interface Select the least used channel C within its transmission range Map C to an interface at node u Map C to an interface at node v Map C to the link (u, v)
Step 8.
Else if node u has all its interfaces assigned If any interfaces at node v is still unassigned Select the least used channel C among those mapped to the interfaces at node u Map C to an interface at node v Map C to the link (u, v) 32
The connected graph is given as input to the algorithm (Skalli et al., 2007) and traverses through each node based on the rank, and finally the interfaces of each node is mapped with channels.
3.5 SIMULATION The centralized rank based CA is tested in Ns-2 network simulator and also the patches of MRMC are applied. The IEEE 802.11 standard is used for simulating the algorithm and the rank based CA performs well in multiple routing protocols. The DSR and AODV (Ad-hoc On Demand Distance Vector) algorithms are combined with CA to test the performance of the algorithm. The bandwidth is set to 2 Mpbs and the traffic types used in the simulation are CBR (Constant Bit Rate), VoIP, FTP and Video-onDemand. The square grid topology comprising of 20 nodes is used. The center node is designated as the gateway and the rest of the nodes are configured as routers. Each node is equipped with two interfaces and the interference range is set to twice as the
Aggregate Throughput(Mbps)
transmission range.
4
WCETT with 5 channels
3.5
WCETT with 2 channels
3 2.5 2 1.5 1 0.5 0 0
10 20 30 40 50 60 70 80 90 100 Time(s)
Fig 3.3. The aggregate network throughput for 10 flows
Initially, ten pairs of nodes are selected, and then CBR User Datagram Protocol (UDP) flow is assigned to each pair. The flow rate is configured between 0 and 0.8 Mbps and five channels are used. The packet size is 1000 bytes and the algorithm is evaluated 33
for 100 seconds. Fig 3.3 shows that aggregate throughput is increased when more number of channels used for transmission, but delay is increased due to channel switching (Fig 3.4). The better throughput is achieved in Rank based Centralized CA method compared to Centralized Quasi-static CA (Ren and Qui, 2009). 6
Delay (ms)
5 4 3 2
1 0 1
2
3
4
5
Number of Channels
Fig 3.4. Number of channel vs. delay
From the simulation, it is observed that the channel 1 overlaps with channels 2,3,4,5 and similarly the remaining channels, 2 to 11, overlapping with other channels as indicated in Table 3.1. Table 3.1. Channel overlap
Channel 1 2 3 4 5 6 7 8 9 10 11
Overlapping Channels 2,3,4,5 1,3,4,5,6 1,2,4,5,6,7 1,2,3,5,6,7,8 1,2,3,4,6,7,8,9 2,3,4,5,7,8,9,10 3,4,5,6,8,9,10,11 4,5,6,7,9,10,11 5,6,7,8,10,11 6,7,8,9,11 7,8,9,10
Non Overlapping Channels 6,7,8,9,10,11 7,8,9,10,11 8,9,10,11 9,10,11 10,11 1,11 1,2 1,2,3 1,2,3,4 1,2,3,4,5 1,2,3,4,5,6
34
3.6 CONCLUSION The typical single channel single radio WMN is not ample for current technological advances and also inadequate for the growing population of Internet users. In this chapter, MRMC WMN is described which focuses on the load in a node by efficiently assigning channels onto the interfaces. A rank based centralized CA algorithm is proposed with the aim of reducing the interference to maximize the network capacity and throughput by estimating the load, distance from global network and number of interfaces in a node. It is concluded that with the raise in the amount of channels used in the channel assignment algorithm, the aggregate network throughput is also proportionally increased.
35
Chapter 4
An Efficient Interference Aware Partially Overlapping Channel Assignment and Routing Algorithm 4.1 INTRODUCTION As, to use the gadgets does not require much expertise, the growth of gadgets in use is extremely increasing in exponential. These gadgets depend on WMNs for Internet access to a large extent. Until the current frequency distribution is sufficient for the needs of all these gadgets, the network keeps servicing the clients without hassles. But will it be able to satisfy this requirement is a question behind, as this network has to host all its clients coming into the span of its location and assuring transparency at the same time. When it becomes a must for all the clients to share the common frequency, interference should not become a factor that disturbs the fine servicing of the network. To keep interference at naught, and to provide a fair bandwidth distribution (Chaudhry et al., 2012), the count of client should be in a limit. The knowledge and skills behind the technology intellectually should overcome the interference problem and provide maximum support for the client to utilize the network and in parallel concentrating and allowing network scaling. As per the IEEE 802.11b/g standards, out of 11 frequency spectrums, 3 nonoverlapping channels of the frequency spectrum are used and the remaining 8 are kept unused. This gives a hint for an intellectual plan to utilize the unused spectrum to overcome interference. With no comprise on bandwidth utilization, maximum performance can be achieved using multi-channel and multi-radio. Multi-radio is achieved by having dedicated NIC to every link into the mesh node. Here, to attain parallel data transmission/reception unique frequencies are set for every link. This chapter concentrates on hybrid MCMR-WMNs. MCMR-WMNs helps in attaining better network connectivity and higher throughput by eliminating the interference. The hybrid WMN is organized into three layers of devices. 1) The lowest layer consists of Wi-Fi, Wi-Max, cellular networks, conventional clients and mobile 36
nodes. 2) The middle layer hosts the router that transmits data between the clients and the gateway nodes. 3) The gateway makes up the highest layer. Interference is a peril behind employing the network bandwidth efficiently and attaining the higher throughput. The interference is a major challenge in MRMC (Kim and Gerla, 2012) which occurs from the parallel transmissions among the neighboring nodes, it will affect the network parameters such as delivery ratio, throughput and capacity. Therefore, the objective of this work is to reduce the interference by efficiently allocating the channels to NICs, thereby enhanced network capacity, better throughput and scaling and also ensures the optimal route to the destination. Raniwala et al. (2004) proposed a centralized load-aware CA and routing, where the gateway acts as a central controller. The physical topology and traffic information are communicated to the gateway. In this chapter, the partially overlapping channel assignment and SINR based routing is proposed, where the central controller does not need to be aware of any traffic information, it just requires a set of active nodes and links. JOCAC algorithms (Rad and Wong, 2006) designate channels to interfaces by estimating the congestion cost on the links and path loss information. JOCAC is presented as a utility maximization problem and it is much suitable for both distributed and centralized CA. Mishra et al. (2006) experiments proved that the network performance and capacity can be improved when orthogonal channels and POC are used together for CA. They have validated that the good spatial re-use of the channels provide enhanced performance. The MMAC protocol is formulated by So et al. (2004) and it permits multi-channel on the end user by adjusting the frequency dynamically using a single interface. This MMAC protocol utilizes only 3 non-overlapping channels, of which 1 channel is dedicated for control packets and 2 channels for data packets. Also, the energy of the node is maintained by combining the power saving strategy with MMAC protocol. Liu et al. (2010) proposed Load-Aware CAEPO (Channel Assignment Exploiting Partially Overlapping Channels), which utilizes the POC along with non-overlapping channels. The nodes are grouped based on their interference range and one group leader 37
is elected in that range which is responsible for assigning channels to the nodes. Khan and Loo (2012) proposed CSROR which provides secure routing and also satisfies the QoS for the real time traffic. Q-SMS routing scheme (Zafar et al., 2012) finds the unused capacity of the link to decide on the suitable admission control policy to ensure the QoS in MANETs. The following sections of the chapter are organized as follows. Section 4.2, elaborates the three types of interferences and its effect in the network. The channel assignment algorithm with POC model using graph edge coloring and SINR routing are elaborated in Section 4.3. The Section 4.4 analyzes the results and performance evaluation using the simulation. Finally, the Section 4.5 is concluded with a summary of findings and the future plan.
4.2 INTERFERENCE Interference is a major threat in wireless networks as it modifies the data and can cause unwanted effect. Simultaneous transmissions from the neighboring nodes within the same interference range increase the interference and diminish the throughput. In the end, the overall capacity of the network is downgraded as a result of the interference. The channel capacity is estimated by Shannon-Hartley theorem, by considering the presence of noise in terms of bps. C B log 2 (1
S ) N
(4.1)
Where bandwidth of the channel is denoted by B (in Hertz) and the signal-tonoise ratio is denoted by S/N. The Shannon-Hartley theorem provides the measures to recognize various kinds of interferences in the link. Let us assume that the nodes P, Q, R, and S are placed close to each other’s interference range; the nodes P and R are senders and the nodes Q and S are receivers. 4.2.1 CO-CHANNEL INTERFERENCE This phenomenon occurs when NICs of two distinct communicating pairs of nodes are assigned to the same channel and are situated close to each other’s interference range, and sending data simultaneously as indicated in Fig 4.1. Let us assume that the 38
nodes P-Q and R-S in Fig 4.1 are assigned to channel 1 and the subsequent events are: Node P has a data to node Q. It uses CSMA/CA to detect if the medium (channel 1) is busy or idle. If the medium is busy, it will wait for the medium to be available through back-off algorithm. In the case of medium being free, the transmission is initiated. When the node P is in the middle of transmitting data to the node Q, and at the same time the node R also wants to send data to the node S, in this case the node R finds the medium in busy state. Hence, the node R applies the back-off algorithm to keep sensing the medium until it is available. Once the medium found to be available, the data transmission started from the node R to S. This kind of interference can easily be detected by CSMA/CA algorithm, so its effect is minimal compared to other two interferences.
R
Q
S
CH1
CH1 6
P
Fig 4.1. Co-channel interference
4.2.2 SELF-INTERFERENCE
CH1
CH1
P
Q
S
Fig 4.2. Self-interference
Self-interference is experienced when a node with two NICs is assigned both the interfaces to the same channel and transmitting at the same time through both the interfaces as indicated in Fig 4.2. Assume that the node P is sending data to the nodes Q 39
and S in Fig 4.2 at the same time, and the node P is built with two NICs. Both NICs are operated at channel 1 and the node P attempts to transmit data in both radios simultaneously. Even though the receiver nodes Q and S are located far from the sender, still the interference will be very severe in self-interference situation. To avoid this kind of interference, the non-overlapping channels should be assigned to NICs of the same node. 4.2.3 ADJACENT CHANNEL INTERFERENCE Adjacent channel interference is caused when NICs of two distinct nodes are assigned to partially overlapping channels as shown in Fig 4.3. Assume that the nodes PQ and R-S in Fig 4.3 are configured to channels 1 and 3, respectively. Initially, the node P begins the transmission, in this case, the node R would sense the medium as idle in channel 3 and it also starts sending the data. As the channels 1 and 3 share the portion of the frequency, the nodes Q and S may not be able to interpret the signal correctly, results in data packet error and it rigorously reduces the throughput.
Q
S
CH3
R
CH1
P
Fig 4.3. Adjacent channel interference
4.3 SYSTEM MODEL The Fig 4.4 shows the design and collaboration among the components in our framework; it provides the detailed information about communication between channel assignments and the SINR computation through routing method. The interference experienced on every link is determined using the SINR computation as part of CA process and the values are persisted into the interference repository.
40
Channel assignment algorithm
SINR Computation
Routing
Interference Database
Fig 4.4. Block diagram that represents the interaction between channel assignment algorithm and the SINR based routing algorithm
Initially, the CA is accomplished and then SINR metric is assigned to the link for finding the optimal path. The CA algorithm can significantly reduce the interference by assigning non-interfering channels and also helps in routing for maximizing the throughput and channel capacity. The CA and routing algorithms combined together and enhanced to increase the overall network performance. The CA, routing metrics and the routing algorithm collectively controls the effect of interference in MRMC WMNs. 4.3.1 CHANNEL ASSIGNMENT ALGORITHM The objective of channel assignment is to expand the number of concurrent transmissions in MRMC WMN, and at the same time reducing the interference between the neighboring links. The graph edge coloring is used to solve the interference problem between neighboring links in channel assignment. The distinct channels are mapped to adjacent links in the graph and all 11 channels are utilized in CA algorithm. The nodes in MRMC WMN is denoted as an undirected graph G = (V, E); where V specifies set of static routers, and E represents undirected links between routers. The CA problem is illustrated as an edge coloring problem of graph theory. Definition 1: Given an undirected graph G, edge coloring algorithm assigns channels to the links of G such that any two adjacent links which are at one hop distance are assigned with orthogonal channels.
41
Strong Edge coloring: Given an undirected graph G, algorithm maps channels to the links of G such that no two links in the neighborhood of utmost two hops distance is configured with same channel. Chromatic Index: It defines the minimum number of channels needed to assign the links of the graph. Consider static mesh routers, each armed with multiple NICs. All nodes in the network are assumed to be with same transmission power. The IEEE 802.11b/g standard is considered for CA and routing and the maximum of 11 channels are used. The routers with gateway functionality are responsible for external access and they hold the complete information and physical connection details about the network, and also builds a physical graph G with a set of active nodes and links A. Assume that nodes u and v require to be configured to channel c; it finds the SINR for channel c for both nodes u and v. The SINR is computed by repeating this task for all the available channels (11). The channel which has less noise or interference is assigned to the interface. The algorithm starts the CA from the gateway node, and then the remaining nodes are carefully chosen according to the distance from the gateway to assign with the channel. In this chapter, color and channel are used interchangeably. Algorithm 4.1: Channel Assignment Using graph Edge Coloring Method
Input: G = (V, E) represent the network V = Set of mesh routers E ∈ V X V is the set of undirected links or edges A = (V, EA) Sub-graph of G is chosen by the CA algorithm. Output: Channels configured to links exist in A 1. Let N = Maximum number of colors or channels (N=11) 2. Let vi = Root of the mesh network for i =1 to K 3. Let L= Number of links or edges incident on vi from A 42
4. For all links or edges e ∈ EA do 5.
Channel(e) = 0
6. while count ≠ L+1 do 7.
for i=1 to K do
8.
Channel (A, G)
Procedure Channel (G1= (V1, E1) G2= (V2, E2)) 1. for i = 1 to K do 2.
if ∃ orthogonal channel c ∈ N , then color the edges of vi
3.
continue
4.
for i = 1 to || unassigned links attached to vi in G1 || do
5.
let l be unassigned links attached to vi
6.
c= Least interference channel not used by links in G2.
7.
c is selected based on signal-to-noise interference calculation
8.
Assign c to link l
9.
if such channel does not exist, then channel with minimum interference is
greedily assigned to link l.
The number of channels needed is computed by the edge coloring method and Vizing's theorem finds it using at most one greater than the maximum degree d of the graph. The coloring problem is NP-complete which is described by Misra and Gries (1992), if d is the maximum degree of the graph, then edges of any graph is colored with d + 1 colors. Let us assume M routers exist in the network for the edge coloring algorithm and the computed SINR values are maintained in interference database. Firstly, the unassigned link is indicated by the value of zero. 4.3.2 COMPLEXITY OF EDGE COLORING The graph in Fig 4.5 is a line graph and it contains no loops. As well the graph has no parallel or multiple edges. It is a plain graph with no pair-wise intersecting edges. There exist a finite number of colors to completely color the connection graph.
43
Facts: 1. Every line graph is claw-free. 2. A r-regular line graph is NP-complete given r is odd. A clique graph can be built by treating each maximal clique as equivalent of a vertex of the graph.
v2
u
v3
v1
Fig 4.5. Connection graph
§ The connection graph in Fig 4.5, the edge coloring problem is NP-complete. This is a simple graph with every vertex of maximum degree 4. The following lemma need to be proved to conclude the NP-completeness of the problem. Lemma: Let G be a simple, connection graph, v be a vertex, v and every vertex including v have degree ≤ 4. If G-v is 4-edge colorable, then so is G. Proof of Lemma: Assume G-v is 4-edge colorable. Define Xi (i=1, 2, 3, 4) as the list of neighbors of v not assigned by color “ i ”. Therefore the 4-edge coloring of G-v is 4
Min | X i | 2
(4.2)
i 1
subject to the constraints | X |2, 1 ≤ i ≤ 4 If G-v is not 4-edge colorable, then 4
| X | 2 * degree(v) – 1 < 2 * 4 i 1
44
(4.3)
4
| X | 8 i & j | X i | 0 & | x j | 4
(4.4)
i 1
Define H Δ(G-v) such that all edges colored by i or j are contained in G. For any vertex not in Xi, the connected component of H containing that vertex must be a path. Interchange the color of Xi and X j | X i | 2 | X j | 2 becomes smaller contradiction
G-v is 4-edge colorable. Without loss of generality, assume Xk = {u} and G’ G obtained by detecting the edge of uv and the edges of color 4. This implies that G’-v is 4-1 = 3 colorable and deg VG’ ≤ 3 and deg VG’ ≤ is 3 colorable by induction. Reconstructing the edges deleted earlier and assigning color-4 to the edge uv, G achieves 4-edge coloring. The Theorem: (G 1) edge coloring algorithm in O (n2) time of the connection graph in question. Proof: Need to show that, for any simple graph G, G (a ) G 1 Where is the chromatic index. Let k = ΔG+1 (N=4 in this case), therefore any vertex of G satisfies the lemma above Edges from G can be eliminated one by one until only one edge if left the resulting graph is K-edge-colorable clearly. If n is the number of vertices, we have ΔG +1, edge-coloring algorithm in O(n2) time. Merging the vertices of degree ≤
G implies Δn= O(m), where m is no of 2
edges the edge coloring algorithm runs in O(nm) time. This concludes that the problem is NP-complete for this connection graph. The SINR is computed at node v using the equation 4.5, and the SINR at receiver v is given by SINR uv
Pu G uv N v Pw G wv w ( u ,v )
45
(4.5)
Where Pu is the transmission power of node u, Guv indicates channel gain for nodes u and v that is influenced by path loss index and the physical distance between nodes u and v, Nv is the thermal noise at receiver v. In this analysis, the chromatic index is 11, because the total number of channels (color) are 11. Figure 4.9 shows the channel assignment in square topology where the problem is considered NP-complete if N = 4. If the nodes are deployed randomly, then there would be a high interference, hence the random topology indicated in Fig 4.8 uses all 11 channels though its chromatic index is 3.
Fig 4.6. Interference aware edge coloring for random topology.
The portion of the random topology is considered to demonstrate the channel assignment using edge coloring in Fig 4.6. The link L1 is tuned to channel 1 and all other links are also within the interference range of link L1, so other links cannot use the channel 1. The chromatic index is 11 in random topology as shown in Fig 4.8, hence Vizing’s theorem proves that strong edge coloring problem is deterministic and colorable by finite number of colors. 4.3.3 ROUTING BASED ON SINR COMPUTATION In wired networks, the shortest path is considered as a path with less number of hops. But in wireless networks, there are many factors that could affect the link quality, so the routing protocol has to include the following constraints in routing metric: maximum bandwidth, minimum hop count, and delivery ratio. The link stability and link quality should be estimated and that can be interpreted as a routing metric for efficient 46
routing. In wireless networks, modeling a new routing metric is ambitious task in view of the following characteristics: Packet loss: If the sender and receiver are separated with larger distance, then multi-hop transmission is required for communication and that would cause fading and packet losses in the links. Finally it would lead to packet retransmission, delay and decrease in the throughput. Packet transmission rate: It is changed dynamically based on traffic condition and also altered in proportion to the current resources such as bandwidth, queue size. Interference: IEEE 802.11b/g standard functioning in 2.4 GHz band which could easily be disrupted by other devices operating in the same frequency band. In addition to that, communication in one link affects the transmission in adjacent link. The intra and inter flow interferences are captured by the routing metric to select the path with less interference. There are two types of interference models applied in the network: protocol interference and physical interference model. The first one is extensively applied to measure the interference and it can be easily applied in theoretical investigation. But the protocol interference model is not better comparing with the later model. The SINR model effectively handles both fading and shadowing for estimating the interference which is typically based on the physical interference model and it is solely depending on the antenna design. The physical interference model is accurate, but it is very complex to use in graph theory based method. The SINR model retrieves accurate value of interference even though it is complex to implement. In this chapter, SINR is presented as a new routing metric to measure interference on a link. The SINR value is interpreted as the value of link cost. The aim of this combined CA and SINR routing is to reduce the interference and total cost of the link, and also balancing the load in proportion to its channel capacity. The interference is detected in the process of assigning channel to NICs; our graph edge coloring algorithm efficiently assigns the channel and reduces the interference. Usually, routing from end devices to the gateway needs multi-hop 47
communication, moreover the gateway is also linked to the Internet which yields backhaul access to the end devices. The physical map of the network and channel information is passed on to the network layer from MAC. The interference database is used to maintain the measured SINR value of the link. The network layer fetches the SINR value from MAC and the shortest path is computed based on the least interference path. 4.3.4 PROBLEM FORMULATION The output of the MAC layer is input to the network layer i.e. CA information is input to SINR routing and the routing protocol computes the optimal route from source to the gateway. A best routing algorithm has to discover an optimal route within a specified time period and it should also reduce frequent update into the routing table to save energy and better utilization of the network resources. Also, it should satisfy Quality of Service specified by the user. In Fig 4.13, the grid topology uses two gateways, so the enddevices can use any one of them. The low interference in the wireless link indicates that the SINR value is high in that link and it is measured through SINR computation. The Shannon’s formula in the equation 4.1 finds the channel capacity which is denoted as Cuv. The data flow from node u to its neighbor node v is denoted as fuv. SINR uv
Pu G uv N v Pw G wv
(4.6)
w ( u ,v )
Where Guv is channel gain and is estimated by a fading model G uv d uv , α is the path loss index, d uv is the distance between nodes u and v and β is the SINR threshold. The cost of link is assessed by the level of noise or interference present on the channel, as a consequence of parallel transmissions and noise. If the SINR value is high, then the quality of the link is considered to be good. Hence, the ultimate aim is to maximize the SINR value and also ensure that the data flow should not exceed the 48
capacity of the link. The SINR routing provides efficient delivery of packets from client to the gateway using interference level as a metric. The SINR routing is denoted as follows: Max SINR uv
(4.7)
u ,v
Subject to
f uv ( u , v )E
f uv ( u , v )E
d u u V
(4.8)
0 f uv C uv
(4.9)
f uv Z
(4.10)
The data rate per second at node u is indicated by d u . The equation 4.8 guarantees the flow preservation at every node. The equation 4.9 (second constraint) indicates that the data flow is less than the channel capacity. The equation 4.10 (third constraint) states that the data flow is an integer value. The graph edge coloring may be sufficient, if we consider only physical topology of the network. But fair and wellorganized routing takes the traffic also into consideration; the second constraint ensures that the traffic should not exceed the channel capacity.
Algorithm 4.2: Proposed SINR routing protocol
Input: Physical topology, active nodes, links and the SINR value from interference repository Output: Optimal route from source to the gateway 1. Read the network topology and set of active nodes and links from the MAC. 2. Assign the cost of all wireless links to zero. 3. Retrieve SINR value from the database. 4. Interpret SINR value to cost of the link and assign the cost to link. Minimum cost of link is β threshold (value 1). 5. Find the shortest path to the gateway (destination). 49
6. If more than one gateway nodes are available, then find the route with a least interference. i.e. Optimal route. 7. If not, then the route is recognized to the gateway. Read Topology and Set of active links from channel assignment
Read SINR value from Database Assign link cost
Find route to gateway node
Is any other gateway nodes
No
Yes For i= 1 to M
Estimate route cost
Estimate optimal route cost
Establish route connection
Fig 4.7. The proposed routing algorithm based on SINR
Table 4.1 indicates the values of SINR, delivery ratio and link cost details. The proposed SINR routing is shown in Fig 4.7.
50
4.4 SIMULATION 4.4.1 SIMULATION RESULTS: CHANNEL ASSIGNMENT ALGORITHM The proposed Interference Aware Edge Coloring problem is experimented in Ns-2 simulation. To support multi-channel multi-radio WMN, the patch is included in Ns-2.33. Different topologies like square, grid and random are analyzed in the simulation and also both OC and POC is given as input for investigation. The flat grid topology in 1000 m × 1000 m area is given as an input dimension. In square topology, the distance between the nodes is 200 m; the transmission range and interference range are 250 m and 550 m, respectively. When the nodes are deployed randomly, the physical distance between any two nodes need not be the same. Each router is equipped with multiple interfaces and the size of the packet is set to 1000 bytes. The data transmission rate is set to 11 Mbps. The signal strength is measured by free space path loss model. The thermal noise power is set to -90 dB and the beta threshold is set to -16 dB. It is observed that optimal performance is obtained at -10 dB. The algorithm is executed for FTP and CBR traffic and with the simulation time of 300 s. Table 4.1. Link cost estimation based on SINR
Value
Delivery rate
Link Cost
SINR ≥ -16 dB
90 - 100 %
1
SINR -10dB to -15dB
79 - 90 %
3
SINR -8dB to -10dB
50 - 79 %
5
SINR < -8 dB
0 – 49 %
10
To validate the CA algorithm, the throughput and the aggregate capacity of the network are assessed. In Fig 4.10, the network throughput vs. the number of NICs is evaluated and it clearly demonstrates that, the proposed POC assignment with SINR routing yields better throughput with superior spatial re-use of the channel. It is evident that the proposed graph edge coloring method provides enhanced throughput with both OC and POC. The network throughput is proportionally increased with more number of 51
NICs and channels. Fig 4.11 indicates the aggregate network capacity in MRMC WMN. The network capacity is increasing radically with the number of channels; this evidently proves that POC raises the overall network performance in WMNs.
Chan
1
2
3
4
5
Chan
6
7
8
9
10 0
Fig 4.8. Random topology
Chan 1
Chan 5
Chan 8
Chan 11
Fig 4.9. Square topology
52
11
6 Channels
11 Channels 45
Throughput (Mbps)
40 35 30
25 20 15 10 5
0 2
3
4
5
6
Number of Radios
Fig 4.10. Throughput vs. number of radios
Fig 4.11. Aggregate network capacity vs. number of channels
4.4.2. SIMULATION RESULTS OF ROUTING USING SINR The grid topology comprises of 16 nodes as depicted in Fig 4.12, considered for CA and routing. The AODV routing algorithm is altered to use the SINR value as a link cost instead of hop count to discover the shortest path. When the source wishes to send through the gateway, initially it checks its own routing table, if there is no existing path, and then it begins the route discovery process. The route request packets (RREQ) are flooded by the process of route discovery. The link cost from the repository, an RREQ identifier, the initiator address, the initiator sequence number, the target address and the target sequence number are contained in the RREQ packet. The link cost is maintained at two levels, the individual 53
link cost and the cumulative cost incurred for the route that the request packet has travelled so far.
G0
1
G2
4
5
6
8
9
10
12
13
14
3
7
11
15
Channel 1
Channel 5
Channel 6
Channel 8
Channel 11
Fig 4.12. Shortest path from end-user to the gateway node
Each RREQ packet is distinctly identified by id of route request and the sender address to avoid flooding of multiple RREQ. Upon receiving the RREQ packet, the target node sends the route reply (RREP) back to the initiator. At the end, the sender transmits messages to the target node on the discovered path. When the link failure encountered, the upstream node immediately broadcasts the Route Error (RERR) packets to the sender node. On getting the RERR packets, once again the sender initiates the route discovery process. Assume that the nodes 0 and 2 are gateways and the clients are connected to access points 14 and 9 as illustrated in Fig 4.12 and the shortest route is indicated in the figure. The shortest path from access point 9 to gateway is 9-8-4-G0 and from access point 14 to gateway is 14-15-11-7-3-G2, these paths are optimal and produce less interference. The typical AODV finds 14-10-6-G2 as 54
shortest path that is based on the hop count, but SINR routing finds the less interference path 14-15-11-7-3-G2. Let us assume that nodes 1, 5, 6, 9 and 14 are sources and gateway is the destination. The following path shows the optimal route: Source 1: 1– G0 Source 2: 6 – G2 Source 3: 5 – 4 – G0 Source 4: 9 – 8 – 4 – G0 Source 5: 14 – 15 – 11 – 7 – 3 – G2
Fig 4.13. Packet delivery ratio vs. number of channels
The performance of CA, the routing algorithm and SINR metric are analyzed by comparing with other MRMC metrics like ETX and ETT. The following network parameters are evaluated to analyze SINR metric:
End-to-End Delay
Packet delivery ratio
Routing Overhead.
55
Fig 4.14. End-to-end delay vs. number of channels
Fig 4.15. Routing overhead
The simulation results of SINR routing are indicated in Fig 4.13, 4.14 and 4.15. The packet delivery ratio is assessed through successful reception of the packet at receiver. In Fig 4.13, the packet delivery ratio (PDR) versus number of channels is analyzed. It is observed that the packet delivery ratio is drastically improving as the number of channel rises and the same PDR is preserved after 4 channels. If SINR metric is used in MRMC WMN, then 90% of the packets are successfully reached at the receiver. 56
Fig 4.14 illustrates how end-to-end delay varies against all 11 channels. End-toend delay includes transmit delay, queue delay and propagation delay, for a packet to travel from sender to receiver. Also, the route request and route reply packet delay is included in this end-to-end delay. Fig 4.15 shows the routing overhead caused by joined CA and routing algorithm. Some of the packets are retransmitted because of packet loss due to the interference, this retransmission delay is included in routing overhead and it is determined by number of retransmission required per flow between sender and receiver. Fig 4.15 indicates edge coloring channel assignment and SINR routing takes less retransmissions compared to ETX and ETT metrics. Hence, the shortest path estimated using SINR interference method is more reliable. Table 4.2. Performance improvement of SINR metric over ETT and ETX
ETT
ETX
Packet delivery ratio
+12.5%
+20%
End-to-End delay
-28%
-41%
Routing overhead
-21%
-30%
The Table 4.2 shows the performance improvement of SINR metric over ETT and ETX metrics. It indicates that SINR metric produces better packet delivery ratio than ETT and ETX metrics, by almost 13% and 20% respectively. The end-to end delay and routing overhead are reduced significantly.
4.5 CONCLUSION This chapter proposed a combined channel assignment and routing in MRMC WMNs that utilizes the POC along with orthogonal channels. The SINR routing metric is modeled for interference. The channel capacity in a noisy condition is measured for data transmission. The simulation outcome indicates that the combined CA and interference aware routing, presented in this chapter, increases the overall network performance and aggregate throughput. It is evident that, efficient channel assignment in partially 57
overlapped channels would increase the overall performance of the network. In addition to that, the SINR metric efficiently captures the interference and delivers high throughput in comparison to other metrics like ETT and ETX. The frequency band of IEEE 802.11b/g is exploited with the spatial reuse protocol to support more number of simultaneous data transmission.
58
Chapter 5
QoS Guaranteed Intelligent Routing Using Hybrid PSO-GA 5.1 INTRODUCTION The emerging technology in wireless networks link individual people and enterprises through Internet and also the society is networked. Various service providers offer Internet facilities to users for web browsing and email applications, but the traditional technology lacks in cost and bandwidth restrictions. The maximum data rate provided by 2G GPRS (General Packet Radio Service) and 3G (Third Generation) networks are 114 kbps and 2 Mbps, respectively. The 2G is developed mainly for voice message and also provides slow data services. The major disadvantage of 2G (Second Generation) and 3G services are decreased timeslot, because wrapping of different services together and it is appropriate for web browsing and email, and not for real time applications like audio and video streaming. For example, watching YouTube videos constantly for 10 minutes takes about 30 to 35 MB roughly and it is proved that the cable Internet access such as cable broadband, DSL (Digital Subscriber Line) and Wi-Fi are required for very high speed Internet amenities. Installing fiber optic cables, building the cellular base stations and maintaining the wired line for service provisioning drastically increases the expense of infrastructure itself. The innovative and cutting edge technology in the field is the advent of WMN which offers high bandwidth and imbibes low cost of deployment and maintenance. WMN is the next generation network that aims at providing the high speed Internet access to any user and features self-configuring and self-healing properties. The major challenge faced by researchers in the development of WMNs is to select an optimal path which avoids interference and also increases the performance. The interference is not only between the neighbouring links which are assigned with the same channel, but also from the adjacent channels and self-interference. An optimal method is needed for routing and the algorithms should also be fast enough to converge for large WMN.
59
The WMN concerns about the capacity, which is achieved through the multichannel and multi-radio, and the characteristics that affects the capacity of the network are bandwidth, interference, delay and jitter. The WMN is rapidly emerging in the recent years due to its potential applications such as community networks, broadband home networks and commercial networks. As these networks require low investment and minimal infrastructure for deployment, they can be used to leverage converting cities into smart cities. It acts as an interface between the users who wants to connect their laptop and smartphone to access applications over the Internet. The biggest challenge in designing a smart city is that achieving the optimal routing capability with efficient resource utilization and at the same time the QoS is also fulfilled. Simultaneous data transmission and reception is possible by routers as they are basically built-in with multiple radios and abundant power. Mesh clients can play the dual role that they can operate as an end-user as well as a router. In this chapter, an example of IEEE 802.11b/g wireless technologies has been considered which operates on the 2.4 GHz spectrum frequency, and the spectrum band is split into 11 channels, out of which 3 channels are non-overlapping. Because of the limited availability of non-overlapping channels, all the nodes were assigned to the same channel and that lead to more performance degradation of the network. When the same channel is assigned to all the interfaces, the performance of the router is more degraded than with the partially overlapping channel assignment. When MRMC-WMN considered along with the partially overlapping channels assigned to the radio on a mesh router, it is observed that the low cost IEEE 802.11b/g mesh networking hardware greatly improves the capacity of the infrastructure mesh networks compared to other existing technologies. When the routers are equipped with multiple radios, the traditional shortest path routing algorithm does not work well and the traditional routing algorithm finds a path without considering the different channels assigned to the radios. In this chapter, an optimal intelligent routing algorithm using Particle Swarm Optimization (PSO) and Genetic Algorithms (GA) i.e. PSO-GA is proposed, which finds the most optimal routes and satisfies QoS constraints.
60
The rest of this chapter is structured as follows: Section 5.2 presents the existing routing algorithms. Sections 5.3 and 5.4 describe the original PSO algorithm and GA algorithm respectively. Section 5.5 presents the hybrid PSO-GA algorithm. In Section 5.6, the proposed QoS intelligent routing, determination of fitness functions, finding of successive particle using operator and crossover operator are presented. Section 5.7 elaborates the simulation and results. The conclusion of the chapter is in Section 5.8.
5.2 ROUTING ALGORITHMS The three classifications of routing algorithms are: reactive, proactive and hybrid algorithms. The proactive algorithms try to find the routes to all other nodes, maintain more than one routing tables and the routing information is updated at predefined intervals. The reactive algorithms are on-demand basis and it finds the route only when it is needed. The AODV algorithms are on-demand based, which does not provide any QoS guarantee. The existing routing algorithm is built around the statistical parameters of a link and the interference is a key factor which influences the packet delivery between the mesh routers. Routing is challenging due to interference and the unpredictable nature of the wireless environment. Routing in WMN is a multi-objective optimization problem when it comes to satisfy the QoS constraints such as bandwidth, delay, interference and jitter. Various kinds of meta-heuristic and natural mechanisms such as ACO, PSO, GA, and neural networks used to find a good solution for a complex problem with fast convergence time. But those algorithms lack to solve the case of routing in MRMC WMN with multiple objective functions subject to multiple QoS constraints. A multi-objective optimization problem is proposed which minimizes the cost while maximizing the channel utilization and also maximizes the network performance while minimizing the interference.
5.3. PARTICLE SWARM OPTIMIZATION PSO is developed by Eberhart and Kennedy (1995), inspired by the social behavior of the species such as bird flocking or fish schooling, used to solve the meta-
61
heuristic optimization problem. PSO is an iterative process which guides to explore and exploit the search space. The groups of entities in PSO are called particles which have a position and velocity and each of the particles explores the solution in the multidimensional search by adjusting the position and velocity. The particle position gives a candidate solution in the search space, and the individual particles have no intelligence and it just follows the simple basic rules in a decentralized manner and acts upon based on the local information. Each particle has memory and the previous state is remembered, the individuality retains the particle’s prior best position and the sociality holds the neighbour’s previous best position. Each of the particles remembers its best value using its own experience; the best value is represented by pbest and the position is represented by pbestX[], pbestY[]. Each particle knows the global best position and it is represented by gbest. The gbest is the knowledge of the group and this knowledge is informed to all the individuals. PSO is the fastest search method for many complicated problems, and the performance of each particle is evaluated based on the fitness functions. At each iteration the particle’s velocity and the position are altered by equation 5.1 and 5.2
vi (k 1) vi (k) s1r1[pi (k) x i (k)] s 2 r2 [p g (k) x i (k)] x i (k 1) x i (k) v i (k 1)
(5.1) (5.2)
Where: i =1,2, .. M; k = 1,2,3,.. t. M is the swarm size and t is the boundary of iteration. pi and gi are the local best and global best solutions; s1 and s2 are cognitive and social learning factors in the acceleration and these values are between 0 and 2; r1 and r2 are arbitrary numbers between 0 and 1; w denotes the inertia weight or weighting functions that stabilizes the PSO algorithm between the local and global search. The largest value of the inertia weight facilitates the global search and the smallest value leads to the local search. The xi(k) is the position of the particle and vi(k) is the velocity of the particle at i-th iteration, and pi(k) and pg(k) represents pbest and gbest. The position of the particle at i-th iteration is represented as xi (k) = (xi1, xi2,... xid) and the velocity is denoted as vi(k) =(vi1, vi2,... vid) in a d- dimensional vector.
62
PSO algorithm for routing: Agent index is ai for an arbitrary i; Particle index is pi for an arbitrary i; Step1. Initialize ai with the position and two velocities randomly. Step2. Find the fitness value of each ai Step3. Calculate pbest and gbest for each agent ai Step4. Do Alter the position and velocity of each agent: vi (t 1) vi (t ) c1r1[pi (t ) x i (t)] c 2 r2 [p g (t) x i (t )] , x i (t 1) x i (t ) v i (t 1) .
Calculate the fitness[ai], the fitness value of each agent If the current fitness value is better than the agent’s pbest Alter pbest of each agent ai Alter the gbest. best value:= gbest. Repeat till the stop criterion. At each iteration, agent searches for the optimal solution by adjusting their properties. The main drawback of PSO is that it easily drops into a local optima due to the fact that the particles rapidly get converged into the best particle. Many improvements and the modification have been introduced on the original PSO algorithm to avoid falling into the local optima.
5.4 GENETIC ALGORITHM The genetic algorithm (Holland, 1975) is a type of evolutionary algorithm which uses the operations, such as selection or reproduction, crossover or recombination, and mutation to produce solutions. The algorithm starts with generating initial population randomly and in each generation, the fitness function is evaluated for every individual; the individuals are picked up from the current generation based on the best fitness value 63
and improved through crossover and mutation operations to create a new population for the next generation. The search process continues until a satisfactory fitness level or finite number of generations is reached.
Start
Initialize population randomly
Find fitness of each individual
Selection Select best individual to pass to next generation
Crossover Recombine portion of two good individual to create best individual
Mutation Flip some of the bits in new individual
No Termination Check
Yes Stop
Fig 5.1. Genetic algorithm
64
The disadvantages of genetic algorithm are:
There is always an issue in finding the fitness function.
Designing the stopping criteria becomes a major issue.
Not applicable for dynamic data, but the network today is more dynamic. Hence, this characteristic is a hurdle to use this algorithm in networking.
Crossover and mutation operations are also difficult to define in the networking.
5.5 HYPRID PSO-GA The hybrid PSO-GA (Settles et al., 2005) integrates the strength of PSO with genetic algorithms, and the hybrid algorithm merges the standard velocity and position update procedures of PSO (Eberhart et al., 1995) along with the objectives of selection and crossover from GAs (Holland et al., 1975). The global search is performed by GA and the local search is carried out by PSO. The hybrid PSO-GA optimization process forages the solution space optimally to reach the gateway. The hybrid approach removes the weakness of PSO and GA, and also the balance of good knowledge sharing and natural selection to provide an efficient and optimal search in the solution space. Algorithm: Step 1. Select the part of the best particle and keep it in a set called elitism. Step 2. If N is the size of particles and Nelitism is the particle in elitism set, select the particles using the following formula to apply PSO rules of the standard velocity and position update. ( N N elitism ) x Breed Ratio
(5.3)
Breed ratio is specified between 0 and 1. Step 3. Apply crossover and mutation operation on the remaining particle. The position of the particle is updated based on velocity propelled averaged crossover (VPAC) method x 'p
xp xq 2 1 v p
, x 'q
65
xp xq 2 2 v p
(5.4)
Where 1 and 1 can be taken within the range (0,1). x’p and x’q are two children created by the particle p and q. xp and vp are current positions and velocities of the particle.
5.6 THE MATHEMATICAL MODEL FOR QOS INTELLIGENT ROUTING In the communication graph, G = (V, E), Where V represents the set of routers and E illustrates links between the routers, the edge lij is between node i and j. The QoS parameters are: the bandwidth BWij, the delay Dij, the jitter Jij and the interference Iij. There are many sources-destinations pair and many possible routes between the source and the destination. The routers are connected to the outside world through gateways and the gateway is responsible for sending and receiving the data. The QoS intelligent routing is used to find an optimal route from the source node to the gateway that cut down the cost and also satisfies the QoS constraints bandwidth, delay, jitter and interference. The problem is to identify a route from the source to the destination that cut down the cost subject to QoS constraints. The objective function f(x) requires the least cost of the path x. Minimum f(x) =
cos t(lij )
(5.5)
lijx
S.t min BWij BWreq
lij x
D ij D req
(5.6)
(5.7)
lijx
J ij J req lijx
(5.8)
I ij
(5.9)
66
Equation (5.6) gives the bandwidth constraints, (5.7) gives the delay constraints, (5.8) gives the jitter constraints and (5.9) gives the interference constraints. The link in the graph has four weights BW, D, J, and C which represent the bandwidth, delay, jitter and cost. The cost of the link is sum of these four weights. is the interference threshold. Transmission on link 1 with channel 1 can be viewed as interference to transmission on link 2 with adjacent channel 2, and the interference is given by I-Factor (i, j) (Mishra et al., 2005). The SINR is modeled as I-factor. 5.6.1 FITNESS FUNCTIONS In QoS intelligent routing, the evolution is determined by the fitness function value which gives the quality of each particle. The fitness function is calculated for each particle in the individual swarms in each generation. The fitness function is determined by summing of the penalty and objective functions. The penalty function determines the degree of penalty for violating the QoS constraints. The penalty function p(x) is determined as:
p(x) 1 max( BWreq BWij ,0) 2 max( Dij Dreq ,0) 3 max( J ij J req ,0) I factor Where 1, 2 and 3 are real numbers used for normalizing the bandwidth, delay and jitter, and these are called punishment coefficients. BWreq, Dreq and Jreq are the values of bandwidth, delay and jitter specified by the application. The fitness function for a hybrid PSO-GA is determined as follows: F( x ) f ( x ) p( x )
(5.10)
If p(x) value is 0, then the QoS constraints are satisfied and the packets are sent through the interference free path, otherwise p(x) is between 0 and 1. 5.6.2 THE HYBRID PSO-GA ROUTING The input to PSO-GA algorithm is specified in the procedure of the particle. The different routes between the source and the gateway are encoded as a particle such as the sequence of nodes is represented as a particle which is encoded as an integer value. So 67
the subtraction between two positions in equation 5.1 and the addition of position and velocity in equation 5.2 are not suitable for this problem. Particles are initialized with a random position and velocity. The breed ratio determines the amount of population, which undergoes PSO or GA. The value of the breed ratio ranges from 0.0 to 1.0. Breed ratio is set to 0.5, so half of the particle is updated by PSO, and the remaining half is updated by GA simultaneously. In Fig 5.2, the flowchart of the hybrid algorithm is represented. Algorithm Step 1. Initialize each particle with a random position and velocity Step 2. Find an initial solution from the source to the destination based on the minimum cost from one node to another and put all the nodes accessed into a set called elitism. Step 3. If N is the cumulative number of nodes and Nelitism is the number of nodes in the elitism set then select the following number of nodes from the set other than elitism set Z ( N N elitism ) Breed Ratio
(5.11)
Step 4. If Z has a decimal value then round it, select Z nodes randomly and update the velocity and position as follows: x i (t 1) x i (t ) c1r1 (pi (t )) c 2 r2 (p g (t ))
(5.12)
xi(t) is the sequence of nodes expressed by a particle, pi(t) is the pbest and pg(t) is the gbest. Step 5. Now the left over nodes other than the elitism set and PSO updated are updated using GA crossover operations.
68
Start
Initialize each particle with random position and velocity
Find elitism set = { path from source to destination based on minimum cost } Nelitism=number of particles in elitism in set N= Total number of particles
Breed_Ratio=0.5 Z=(N-Nelitism)*Breed_Ratio
For Nelitism particle selected do the following
For Z particles selected do the following
Crossover using VPAC operator
Update pbest
No
Update gbest Stopping
criteria? Update velocity
Yes Optimal Path Update position Stop
Fig 5.2. The hybrid PSO-GA routing algorithm
69
Operator Assume that pi= (x1, x2, x3 , …, xk) and pg=(y1, y2, y3,…, yk) For example if pi = (1, 2, 4, 9, 13) and gi = (1, 7, 5, 10, 13) Pa’ = pi pg Pa’= {1, alter (2, 7), alter (4, 5), alter (9, 10), 13} Pa’= {1, 7, 5, 9, 13} Alter (2, 7) = min {(s, 2), (s, 7)}, where s is the source node. Find the minimum cost of the source to node 2 and 7. Node 7 is having the minimum cost path from the source node, so the node 7 is included in the set and the node 2 is eliminated. Repeated nodes in Pa’ is eliminated. Crossover operator Two particles are selected randomly from the population for a two point crossover. Two points are selected for the crossover, a sequence of nodes from the beginning of the particle to the first crossover point is selected, the part of particle from the first point to the second point is selected from the second particle, and the remaining is copied from the first particle. For example P1 = {1, 7, 5, 8, 12, 15, 21, 24, 25} P2 = {1, 7, 5, 10, 17, 19, 22, 25} P1’ = {1, 7, 5, 10, 12, 15, 21, 24, 25} P2’ = {1, 7, 5, 8, 12, 15, 21, 25} Sometimes the crossover operator and operator lead to unconnected route, So that we need to be careful while finding the fitness values of this particle. At each iteration there exist two or more redundant particles. These duplicate particles are discarded at each iteration to increase the searching ability.
5.7 SIMULATION Java and JADE framework is used to simulate the QoS intelligent routing algorithm. The network topology taken for the optimal routing is shown in Fig 5.3, 70
comprises of 25 nodes. The node 1 is the source node and there are 3 gateways to connect to the Internet, which are node 11, 13 and 25. The performance of the intelligent routing algorithm is tested for 50, 75, 100 and 125 nodes. Each node is equipped with multiple network interfaces which are tuned to multiple channels. Many possible routes are available between the source and the gateway when the network size is larger or it is densely connected.
4
9
13
16
19 22
2
18
17
10
5 1 7
25 21
15
12 8
3 6 11
14
20
23
24
Fig 5.3. Random topology
The physical distance between any two nodes differs randomly. The transmission range of nodes is set to 250 meters. The link exists between the two nodes if it is within the hearing range of each other. The link cost is specified in the cost matrix within the range of [2-10] and the bandwidth is set to 11 Mbps uniformly for all the links. Similarly, the delay matrix within the range of [0.5 ms - 2 ms], the packet loss matrix in the range of [0.001-010] and the jitter matrix is in the range of [0.5 ms - 2.0 ms]. The interference value is normalized between 0 and 1, and assigned to each link which is specified in the I-matrix. The BWreq, Dreq, Jreq values differ from application to application. The different source and destination nodes are selected for various runs in the same test.
71
5.7.1 SIMULATION RESULTS Fig 5.4 depicts that the value of fitness vs. number of iterations for 25 nodes. The performance of PSO, GA, hybrid is evaluated, at 14th iteration; Hybrid PSO-GA gives an optimal path whose fitness is 12.56. Fig 5.4 shows the progress of algorithm finding the optimal path for topology given in the Fig 5.3. PSO finds the optimal path at 20th iteration, but GA gets 16.56 at this iteration, which is a global optimal route, so GA needs some more time to converge. The Table 5.1 indicates the path taken and the fitness value at each iteration.
Fig 5.4. Number of iteration vs. value of fitness
Fig 5.5 shows the computation time for PSO, GA, hybrid with increase in the number of nodes in the network. It shows that the computation speed of all three algorithms reduces, when the network is extended. The hybrid algorithm shows better performance compared to PSO and GA, it yields the optimal solution quickly, when more than 100 nodes added to the network. Fig 5.6 shows the packet delivery ratio vs. number of nodes. The hybrid gives better performance compared to PSO and GA, the packet delivery ratio is retained at 90% even after the network size reaches 100 nodes. It shows that hybrid approach guarantees the QoS and is more suitable for reliable communication in MRMC-WMN. 72
Table 5.1. Path taken at each iteration and fitness in hybrid PSO-GA algorithm
Iteration 1 2-4 6-10 10 11-13 14-20
Path 1-7-5-10-13 1-7-8-12-11 1-3-6-11 1-2-4-9-13 1-2-4-9-13 1-7-5-9-13
Fitness value 20.9 20.34 18 16.56 16.56 12.56
Fig 5.5. Number of nodes vs convergence time
Fig 5.7 shows average end-to-end delay vs. number of nodes. The delay escalates steadily when the number of nodes in the network increases. The hybrid approach gives a smaller average end-to-end delay compared to PSO and GA. The hybrid model outperforms the PSO and GA in terms of the convergence time, packet delivery ratio and the average end-to-end delay. Thus, it is evident that the hybrid PSO-GA is very suitable for optimization of routing in MRMC-WMN with POC assigned. Other techniques like PSO and GA fail to find the optimal solution in a large WMN with multiple constraints. 73
Fig 5.6. The packet delivery ratio vs. number of nodes
Fig 5.7. The average end-to-end delay vs. number of nodes
74
Table 5.2. Performance improvement of Hybrid PSO-GA over PSO and GA
PSO
GA
Convergence time
-28%
-39%
Packet delivery ratio
+5%
+7%
Avg. End-to-End delay
-6%
-12%
The Table 5.2 demonstrates the performance improvement of Hybrid PSO-GA, where 28% and 39% decrease has been recorded in convergence time compare to PSO and GA, respectively. The packet delivery ratio in Hybrid PSO-GA outperformed PSO and GA, and the average delay is decreased to 6% and 12% than PSO and GA, respectively.
5.8 CONCLUSION The QoS guarantee is essential for real time communication, but it is hard to achieve QoS in wireless networks. In this chapter, an intelligent routing using the hybrid PSO-GA is proposed to support QoS. The hybrid algorithm removes the weakness of PSO and GA, and it increases the stability between the knowledge sharing and the natural selection to find the optimal solution in the search space. Half of the particle is updated by standard position and velocity update of PSO and the remaining half is updated by crossover operation of GA simultaneously. The QoS parameter and the interference value are added into the fitness function to find the optimal path. The simulation results indicate that the hybrid algorithm efficiently resolves the QoS routing, also gives less convergence time, end-to-end delay and the better delivery ratio.
75
Chapter 6
Coefficient of Restitution based Cross Layer Interference Aware Routing Protocol 6.1 INTRODUCTION Ever since the evolution of communication began, QoS has become imperative to be considered in computer networks. Nowadays, multimedia communication on the Internet has been a dominant communication. When the number of users on the multimedia communication channel is increased or more traffic on the Internet, there may be packet loss and quality degradation. The emerging interactive applications like multimedia streaming and multiplayer games demand less round trip time, so RTT plays a significant role in enhancing the throughput. The primary focus is to scale down the RTT, loss rate and collision, caused by interference, for refining TCP performance in WMN. Most of the existing system design considers POC as a danger because it severely affects the transmission between the nodes. An efficient channel assignment technique with POC solves the interference problem and also produces significant improvement in parallel transmission and throughput. The primary focus is on POC, which increases the number of users accessing the Internet. But the major problem with POC is that the interference between the adjacent channels and its effect, as it reduces the network throughput badly. 6.1.1 CONGESTION IN WMN The densely deployed nodes in IEEE 802.11 WMN can cause network congestion that leads to a packet drop, delay in delivery and frequent disconnection. Generally, the data from the source is reached in the Internet through the gateway with multi-hop access, so congestion occurs more near the gateway, but random at other network destinations. Most of the congestion control algorithms in the wired networks try to estimate the available capacity, i.e. bandwidth, queue size in the router, to fix the congestion window on the sender side. These congestion control algorithms do not 76
apparently find the real congestion status of the wireless networks because of various reasons, such as channel interference, mobility and congestion. The various algorithms (Cai et al., 2013; Wang and Perkins, 2008; Al-Jubari et al., 2011) have been proposed for wireless networks for improving the ability of TCP to judge the congestion status more efficiently. Masri et al. (2014) presented NICC scheme for congestion control that considers the neighborhood information for controlling the transmission rate to guarantee unbiased bandwidth allocation. The congestion aware routing proposed by Bhorkar et al. (2015) estimates the congestion measure metric for every neighbor node and the next hop selection is based on this metric. These solutions are categorized into two kinds:
End-to-End congestion control method: It reacts very slowly in wireless networks because of the waiting time for acknowledgement (ACK) is more.
Hop-by-Hop congestion control method: It reacts quickly to detect the status of the link and intermediate nodes, and it can make effective decisions. The hop-by-hop delay is accumulated into the end-to-end delay, so controlling the
single hop delay ensures that the less amount of end-to-end delay. Based on the channel access at each hop, the per-hop delay would significantly change. In this chapter, a Coefficient of Restitution based Cross layer Interference aware Routing protocol (CoRCiaR) is proposed to improve TCP performance in WMNs. The RTS/CTS scheme at the MAC layer is used to estimate the congestion status of the link and a contention mechanism algorithm is proposed at the MAC layer, and then hop-byhop RTT is estimated for dynamic routing. The performance of the algorithm is evaluated using the Coefficient of Restitution (COR). The channel busy time and throughput is considered to measure the network, whether it is highly congested or not. The simulation results illustrate that the CoRCiaR protocol can yield less delay, good throughput and less packet loss to the interference situation. It can also provide QoS and minimize the RTT along the path. This chapter is organized as follows: The system model is elaborated in Section 6.2, in which, contention algorithm in MAC layer is modified and the resultant RTT calculation is presented. The routing algorithm is explained in Section 6.3. The 77
simulation settings, graphs and performance evaluation using COR are analyzed in Section 6.4. The chapter is concluded and the scope is discussed in Section 6.5.
6.2 SYSTEM MODEL 6.2.1 CROSS LAYER APPROACH There are two types of cross layer approaches: loosely coupled and tightly coupled. The parameters in one layer are cascaded to another layer in the loosely coupled method. For example, the interference level in MAC layer is intimated to the network layer. Two or three layers combined into a single layer in the tightly coupled method. For example channel assignment and routing is optimized into a single layer (Prasad and Kumar, 2013).
Routing QoS Based Route Selector Channel Interference, Congestion
Update Routing metric due to new RTT from MAC
MAC
Fig 6.1. Cross layer design
Most of the current protocols are insufficient for handling the cross layer interaction. Wireless mesh networks need more interaction between the layers, such as MAC and routing layers, routing and transport layers. In this chapter the loosely coupled cross layer approach has been used. In this proposed approach, the MAC layer passes the channel interference and congestion information to the network layer, so that the network layer reroute the packet into the congestion free area. The cross layer based hop-by-hop approach dynamically monitors the status of the link at the MAC layer and the status is updated to the network layer to find the congestion free path. The Fig 6.1 shows that the interaction and parameter passing 78
between MAC and routing layers. The MAC layer measures the congestion status, on the basis of contending channel interfered with ongoing transmission of neighboring nodes. The proposed hop-by-hop cross layer approach uses the RTS/CTS protocol for explicit information exchange. 6.2.2 MODEL AND MOTIVATION When a ball is dropped on the floor, it bounces back, but the ball will not reach its starting position. It is a classic problem in physics. The ball’s behavior is identical of a sphere-shaped spring. When the ball hits the floor, it applies a force on the floor greater than its weight, and the floor applies an equal force back. The ball is compressed by this force and the gravitational force. Hooke's law is satisfied for small compression. The gravitational potential energy of the ball before the drop is transformed into kinetic energy and eventually into elastic potential energy when the ball is compressed. Some of the energy is converted into thermal energy by internal friction, as the ball is not perfectly elastic. The thermal energy is not converted back. The ball does not reach its initial height, due to its initial gravitational potential energy is transformed into thermal energy. The phenomenon of “energy loss” is characterized by the COR, the ratio of the speed of the ball after bounce to the speed of the ball before bounce. A perfectly hard floor is a stationery floor, incapable of moving itself. The “stationery behavior” is noted, further. The definitions below are significant in the context. Coefficient of Restitution =
Re bound Speed Incidence Speed
V2 KE rebound rebound Coefficient of Restitution2 KE incidence V 2 incidence
(6.1)
(6.2)
The network is assumed like a gravitational field, the packet is viewed like a ball, moving from source to the gateway, sending a packet and receiving acknowledgement can be viewed as a bouncing ball. The movement of the packet is decided by the gravitational force field.
79
WMN, nodes are stationary, analogous to the perfectly hard floor. The loss of height could be translated to different path lengths a message may traverse, which is due to the loss of energy explained above. Dynamic routing can be interpreted as energy transfer between nodes, i.e. a persistent interaction among the nodes such that the messages are transmitted. A good enough measure of energy transfer is explained by kinetic energy, the definition of which is well known. Let us consider two objects: object 1 and object 2, and they are colliding with each other, in this case, the COR is denoted by (V 2 V1) ( U1 U 2)
COR
(6.3)
Where: V1 is the final speed of object 1 after impact V2 is the final speed of object 2 after impact U1 is the initial speed of object 1 before impact U2 is the initial speed of object 2 before impact. The COR is considered in evaluating the performance of the CoRCiaR approach. In the proposed approach, each node in the network is assigned with gravitational potential V(v), and the interaction (transmission) between the nodes in its vicinity is influenced by force. Let us consider that the packet p in node v is forwarded to the neighbor node to reach the gateway g. The next hop neighbor is identified through the potential field difference between node v and other neighbors. Assume that w is the neighbor of v, here the force is defined as F( v, w ) V( v) V( w )
(6.4)
In this chapter, the force is interpreted as delay and the packet p on node v is forwarded to the neighbor node which is having a minimum delay or force F (v, w). If the node v chooses the node w as next hop rather than node u , then it must hold 80
F( v, w ) F( v, u )
(6.5)
The coefficient of restitution measures the elasticity of collisions. The COR value is 1 for perfectly elastic collision and kinetic energy is well-maintained and multi-hop transmissions may take place. The COR value is 0 for perfectly inelastic collision. The pair of object with zero COR, stops bouncing at all and it implies no transmission of messages. 6.2.3 LENGTH The length (distance) is estimated to find the shortest distance between the sender and the gateway. Each packet is transmitted towards the gateway on the basis of the length field. We define the length field as: Vlg ( v) length( v)
(6.6)
Where length(v) gives the total length of the node v to the gateway. The length(v) is the shortest path which is calculated by considering the RTT as routing metric, So length(v) will have a less RTT value. The distance or length between the node v and the node u, specifically Vlg (v, u), is represented in ms. The length field Vlg (v) is time-based and it dynamically changes when there is any change in the Internet traffic. When the node v has more than one neighbor with different RTT values, then the node v chooses the node with less RTT value as the next hop node. In this fashion, every node calculates the length(v) to discover the list of neighbors towards the gateway, and the nodes maintain a routing table, which holds immediate neighbor and its RTT value. In WMN, redundant paths do exist, so the proposed method considers all the nodes and all the possible routes to discover the congestion free path to route the packets. 6.2.4 MODIFIED RTS/CTS MECHANISM The objective of finding the congestion status at MAC layer is to avoid the packets moving to the interference area. In this approach, a node selects any of its neighbors to forward the packet by inspecting the channel interference. Specifically, a node selects one of its neighbors with less interference towards the gateway, as a next 81
hop node and it transmits the packets in interference free path. Moreover, the congestion at MAC layer is primarily caused by co-channel interference, self-interference and partial channel interference. The congestion status of the link is evaluated based on the RTS/CTS protocol in IEEE 802.11. In this section, the modified RTS/CTS mechanism is proposed, and the following assumptions are made:
The MRMC WMN with 11 channels available for use and the data transmission rate is same for all the channels. Since the channels overlap with each other, transmission in one channel interferes with another channel.
Each router is equipped with multiple transceivers and assigned to different channels. So the router can simultaneously send and receive on different channels at the same time. For example, let us denote two nodes: node1 and node2.When node1 has a data to
send to node 2, the node 1 and node 2 exchanges the RTS and CTS packets to reserve the idle channel. The Preferable Channel List (PCL) table is maintained by each node (So and Vaidya et al., 2004) and it contains the list of desirable channels, which helps in avoiding the interference. The level of preference is divided into three categories:
High preference: The channels that have already been selected by the node in the current beacon interval and each node will have at most one channel in this state.
Medium preference: The channels that are yet to be taken by the node or neighbors within node’s transmission range.
Low preference: The channels that have already been taken by at least one of its neighbor within node’s transmission range. The node 1 prepares to send a packet to the node 2 and it selects the channel c1.
The node 1 is configured with channel c1 and it sends RTS packets to node 2. The node 2 examines the channel c1, to check if any interference with ongoing transmission in node 2. 82
Algorithm 6.1 describes the RTS/CTS method for QoS guaranteed application. When node 1 wants to send a packet to node 2, firstly, the node 1 has to carefully select a channel which is not interfering with other neighbor nodes. The node 1 uses the CSMA/CA to detect the co-channel interference, to identify if the medium is busy, and then the node 1 tries with the back-off algorithm. But, the adjacent channel interference is not detected easily and dealing with the same is important as it would decrease the throughput dramatically. i - Number of interfaces at node 2. c[i] - Assigned channel number at node 2
Algorithm 6.1: RTS/CTS for QoS guaranteed Application for j = 0: i If c1 equals c[i] then Defer transmission else if c[i] equals to channel 1 to 6 then If (c1 = = (c[i] +5)) then There is no interference and no congestion in the channel Send CTS else Defer transmission else if c[i] equals channel 7 to 11 then if ( c1= = (c[i]+ 5)) mod 11 then There is no interference and no congestion in the channel Send CTS else Defer transmission else Defer transmission i i+1 End for 83
Algorithm 6.2: RTS/CTS for delay tolerant Application
for j=0: i If c1 equals c[i] then Defer transmission else if c[i] equals to channel 1 to 6 then If (c1 = = (c[i] +4)) then c1 is overlapping partially, so less interference and it is suitable for application tolerating packet drop Send CTS else Defer transmission else if c[i] equals channel 7 to 11 then if ( c1= = (c[i]+ 4)) mod 11 then C1 is overlapping partially, so less interference and it is suitable for application tolerating packet drop Send CTS else Defer transmission else Defer transmission i i+1 End for
The node 2 has to verify if c1 is interfering with the channels assigned to other radio. If c1 value is matched with any of its interface channel number, then it is selfinterference, so node 2 rejects the transmission. If c1 is mutually orthogonal to already assigned channel number in node 2, then there is no interference and no congestion in the channel. Hence the node 2 sends CTS to node 1.
84
Algorithm 6.2 describes RTS/CTS method for delay tolerant application. The channel separation between c1 and other interfaces of node 2, and its channel number happens to be 4, and then it is partially overlapping channels in the link with less interference. This is suitable for the application which is capable of tolerating delay and packet drop. In MAC layer, the logical status of the link is the congestion, but in TCP layer, if the buffer is occupied, then it is regarded as physical congestion. The RTS/CTS exchange eliminates the packet collision due to the channel interference as well as the over saturation of the MAC layer. The performance degradation of TCP in WMN is mainly due to contention delay affected by RTS/CTS mechanism. 6.2.5 CUMULATIVE RTT The delay comprises of three components: propagation delay, transmit delay and queue delay. But in many situations, we are interested in calculating only the total time it takes to transmit a packet from the sender to the receiver and to receive the ACK back. This is regarded as RTT. Assume that a node v accepts a packet from node u, and the node v does not always select the same node to forward the packet. According to the traffic condition, the delay between two nodes may change dynamically, that result in the same node is not being selected as a next hop. In this approach, hop-by-hop RTT is estimated, in other words, the delay between the neighboring nodes are individually measured and then cumulative RTT is taken at the sender node. The hop-by-hop delay consists of three components: queue delay, contention delay and transmission delay.
Queue delay: The time interval between the packets reaches the queue and moves to the head of the queue.
Contention delay: The time interval between the packet at the head of the queue and to gain access to the physical channel through the channel access mechanism RTS/CTS.
85
The queue delay and contention delay are depicted in Fig 6.2. The contention delay in WMNs with multiple radios is significantly higher compared to the wired network. Hop-by-Hop delay= queue delay + contention delay + transmission delay
ti
Queue Delay
th
(6.7)
ti+1
Fig 6.2. Delay in queue.
Let us assume that the packet size is fixed for all the transmission, so the transmission delay does not change dynamically. The queue delay is primarily determined by the contention delay which is the dominant portion of the total hop-by-hop delay. For each frame, the variables ti, th, ti+1 are maintained to store time components. The variable ti is used to hold the arrival time of the frame at node i, and th records the time at which the frame reaches the head of the queue. The ti+1 record the time at which the frame is transmitted to the physical medium of node i. The time difference between th and ti is called as queue delay and the time difference between ti+1 and th gives the contention delay. Queue delay t h t i Contentiondelay t i1 t h
(6.8) (6.9)
The function Q(v) defines the queue delay at node v. The Q(v) defined as Q(v) (t h t i ) (t i1 t h )
86
(6.10)
The two potential fields, queue delay and contention delay, are the key features of this approach and are used in making the routing decision. For simplicity, queue delay and contention delay are combined linearly as follows: Q(v) (1 )(t h t i ) (t i1 t h )
(6.11)
Where 0 ≤ α ≤ 1, if the value of α is zero, then there is no contention delay, and only queue delay at the node. If the value of α is one, then there is no queue delay, but contention delay at the node. If the value lies between zero and one, then both queue and contention delays at the node. The parameter α controls the degree of influence of two potential fields for making routing decision. Cumulative RTT at node V Vrg ( v) min 0 RTT ( v) n
(6.12)
where 0 RTT ( v) is the cumulative RTT from node v towards the gateway g. n
Here n is the hop count from source to gateway g. The cumulative RTT gives the congestion towards the gateway. Each node in the network sends a packet to the immediate neighboring node to find out the hop-by-hop RTT and updates its own routing table. The sender node compares RTT value received from all of its neighbors, and chooses the next hop with less RTT value, and then finds the cumulative RTT towards the gateway using the equation 6.12. Each node in the network recursively doing this process, so it can determine the congestion and then make a decision to select the next hop.
6.3 ROUTING ALGORITHM When a node is ready with packets to be sent, it first sends RTS to check if the neighboring node is not congested; in case the neighbor is congested, then the sender waits for some amount of time. Once the sender receives CTS, it starts sending the packets to the neighboring node and subsequently waits for the acknowledgement to calculate the RTT value. The RTT value is influenced by many parameters such as: the
87
rate at which data is transferred from the source, the medium used for the transmission (i.e. a wireless, optical fiber or copper), the distance between the source and neighboring nodes, the presence of noise in the circuit, the number of other requests pending at the intermediate nodes, and the speed at which the intermediate node functions. RTT estimation can be used in routing algorithms for calculating the optimal routes. For every hop, sampleRTT is calculated by the difference between the packet sent time and ACK received time. The sampleRTT may vary from packet to packet due to dynamic nature of the channel. To estimate the actual RTT, the average value of sampleRTT is calculated and the AverageRTT (Jacobson, 1988) is estimated as Difference sampleRTT AverageRTT AverageRTT AverageRTT ( Difference)
(6.13) (6.14)
Where δ is between 0 and 1. Since the wireless topology changes dynamically, each node should be able to learn the routes quickly. If any of the nodes are inactive, then the protocol excludes them from the path. So, the hello messages are used by the nodes to indicate activeness and inactiveness to its neighbors. The nodes which are active respond quickly to the new route requests. Hence, there is a need for on-demand routing, which can be achieved using the AODV algorithm. The AODV considers the hop count as a routing metric to find the shortest distance between the sender and the gateway, which does not account the interference on that path. In order to reduce the interference, AODV chooses the routes by keeping RTT as a metric. In Fig 6.3, the mesh topology where the route setup is based on the hop count as a metric, and with a typical AODV approach, the source transmits a RREQ to the destination node (gateway). The route request from node 5 reaches the destination node 4 through path p1 (5-4) faster than through path p2 (5-7-6-4). Since the number of hops is less in path p1, p1 is selected even though more interference on that link. As the selected channels in path p1 are having a high packet drop, it is necessary to dynamically monitor 88
the delay and accordingly select the path by considering the current channel quality, to reroute the packets.
4 1
6 2 5
7
3
Fig 6.3. Routing in mesh architecture
Fig 6.4 explains how the CoRCiaR protocol is performed using RTT as a metric. Initially the route discovery module finds the shortest path based on the number of hops between the source and the gateway. Packets are sent through the shortest path using the typical AODV algorithm and then AverageRTT is estimated for each hop in the network. The values of RTT are sorted and the routing table is re-constructed by replacing RTT as its link values. Again, the route discovery module rediscovers the congestion free alternate path and the new throughput is obtained from the network; this new throughput and the older throughput are analyzed to compare the performances. Routing Algorithm Step 1. Find the shortest path from the source to the gateway (for i = 1 to n ). Step 2. Perform routing using AODV algorithm Step 3. Compute the value of the throughput and delay based on the processing of the packets. Step 4. Find the RTT for each hop Step 5. Sort the values of RTT 89
Step 6. Convert the lower RTT value of each hop to the link cost and assign the cost to the link. Step 7. Perform the routing with new metric Step 8. Calculate the throughput again for the routing done in Step 7.
Read the topology and links between nodes
For i=1 to n
Perform the routing using AODV
Calculate the throughput, Delay, RTT and packet drops for each hop
Assign RTT as a link cost
Perform routing using new metric, which gives the congestion free path
Calculate the throughput and evaluate using COR
Fig 6.4. Routing using RTT as a metric
6.4 SIMULATION The performance of CoRCiaR protocol is evaluated using the Ns-2.29 simulator with MRMC patches included. The simulation uses AODV for dynamic routing and modified RTS/CTS protocol at MAC layer. The nodes are deployed randomly in a 1500m x 800m area for evaluating the performance in chain and random topologies. In random topology as the name suggests, the distance between the nodes are random, wherein the 90
chain topology has the fixed distance of 150 m between the nodes. The transmission range is set to 250 m, and the interference range is set to 550 m. The default data rate 1 Mbps is used and the packet size is set to 1000 bytes. The traffic types used in the simulation is tcp and the simulation is performed for 500s. The comparison study is performed between the CoRCiaR with TCP-AP (TCP with Adaptive Pacing) (ElRakabawy and Lindemann, 2011) and semi-TCP with ACK. For simulation, the network is organized with 20, 40, 60, 80 and 100 nodes, randomly distributed in a flat grid area. Table 6.1. Simulation Settings for CoRCiaR algorithm
Parameters
Values
Platform
Ns2 version 2.29 with MRMC patch
Network Area
1500 m X 800 m
Propagation model
Two ray ground model
Network Topologies
Chain topology and Random topology
Transmission Range
250 m
Interference Range
550 m
Frequency
2.4 GHz
Traffic Type
TCP
Channels
1-11
Packet Size
1000 bytes
Maximum queue length
50
Simulation Time
100s
Transport Type
TCP
Data Rate
1 Mpbs
6.4.1 EVALUATION CRITERIA
Throughput: The throughput is measured at the gateway, and it is obtained by averaging out all the flows at a given time.
91
Number of packets successfully Throughput
received by the gateway Number of packets successfully
(6.15)
sent by the source
End-to-End Delay: The cumulative delay, the packet to traverse, from source to destination nodes. It includes queue, propagation and transmission delays.
RTT: It is the time taken by a packet to reach destination plus ACK back to the source node.
COR: It is the ratio of throughput, before and after the collision at MAC layer. COR
Throughput after drop due to collision at MAC Throughput before impact of collision at MAC
(6.16)
To evaluate the performance of the routing algorithm, the simulation of two existing congestion control methods, semi-TCP with ACK, and TCP-AP, are performed. From Fig 6.5, it is evident that the increase in the path length i.e. number of hops, also increases the RTT values. The Fig 6.5 depicts RTT values for all the three schemes; the x-axis indicates the number of hops, while the y-axis denotes RTT values in milliseconds. The graph shows that the proposed scheme outperforms other two approaches with the clear advantage of predicting the traffic condition and interference at each hop. In the conditions like nodes deployed at random fashion and the network with high interference, the proposed method yields significantly less delay.
Fig 6.5. RTT against number of hops
92
Fig 6.6. Delay vs. number of hops
From Fig 6.6, it can be observed that the proposed method drastically reduces the packet delay compared to the other two methods. Fig 6.7 show cases the throughput obtained by semi-TCP with ACK, TCP-AP and CoRCiaR. The CoRCiaR performs well even with an increased number of hops. The throughput decreases dramatically when the number of hop increases and this is due to channel sharing in the MAC layer. The throughput of CoRCiaR protocol is stable, when the number of hops reaches 4 or more. The other two algorithms obtained lower throughput than the CoRCiaR protocol as the number of hop increases.
Fig 6.7. Throughput vs. number of hops
93
Fig 6.8. RTT vs. number of nodes
RTT is increased when the number of nodes deployed in the network is high. Fig 6.8 shows that CoRCiaR protocol gives less RTT value compared to SemiTCP and SemiTCP-AP. Fig 6.9 shows that the throughput of CoRCiaR protocol is higher than the other two methods.
Fig 6.9. Throughput vs. number of nodes
94
Table 6.2. Performance improvement of CoRCiaR over semiTCP-AP and semiTCP
semiTCP-AP
semiTCP
RTT vs. number of hops
-46%
-44%
Delay v. number of hops
-51%
-65%
RTT vs. number of nodes
-5%
-14%
Throughput vs. number of hops
+47%
+78%
Throughput vs. number of nodes
+10%
+15%
It is observed that CoRCiaR protocol delivers greater improvement in performance than semiTCP-AP and semiTCP. The Table 6.2 shows that a significant decrease in RTT and delay compared with semiTCP-AP and semiTCP. The throughput vs. number of hops can be improved by 47% and 78% than semiTCP-AP and semiTCP, respectively. 6.4.2 PERFORMANCE EVALUATION USING COR In wireless networks, throughput depends upon the packet drop and the whole network performance is determined by the throughput which is calculated using the COR. The COR is the ratio between the throughput derived using typical AODV and the throughput derived through CoRCiaR approach. The throughput is inversely proportional to the RTT value. The COR values lie between zero and one, indicates the elasticity of the collision. If the COR value is 1, then no packet drops in the network and this condition are known as perfectly elastic collision, which produces the consistent improvement of throughput in the network. If the COR value is 0, then significant packets have been dropped and this is known as inelastic collision, in which the performance consistently decreases. When the COR value ranges between 0.0 and 1.0 , few packets drop are seen in the network, which
95
results in consistent improvement of throughput in the network and the same is called as partially elastic collision. Table 6.3. COR values
COR
Throughput of SemiTCP
Throughput of
with ACK(Kbps)
CoRCiaR (Kbps)
483.133
483.133
1
240.936
240.936
1
154.658
177.829
0.869701
101.137
150.523
0.671904
84.6593
147.935
0.572274
75.6836
140.5726
0.538395
57.1282
140.5449
0.406477
56.5467
139.8836
0.404241
47.2099
138.7724
0.340197
47.5675
135.6574
0.350644
48.9261
133.5736
0.366286
50.8329
130.2198
0.390362
48.2715
129.3132
0.373291
48.1595
128.3545
0.375207
47.8169
122.8472
0.389239
45.1668
120.2656
0.375559
48.5566
118.7433
0.408921
46.7605
117.8355
0.396829
49.2422
115.1323
0.427701
Table 6.3 shows the throughput of SemiTCP, CoRCiaR and COR values. It indicates that the proposed method produces the higher throughput than the other two algorithms. So, the COR values are used to evaluate the performance of the network depending upon the values. 96
6.5 CONCLUSION WMN is considered as one of the most reliable and low cost network to provide broadband Internet access. The congestion control in MRMC WMN is different from the traditional congestion control. In this chapter, the proposed CoRCiaR protocol reroutes the traffic in the congestion free path in WMN and the RTT in each hop is considered in making the routing decision. In multi-channel, the adjacent channel interference is very severe, so there would be a significant amount of packet loss and that results in performance degradation. Some modifications in the RTS/CTS scheme can significantly improve the throughput. The proposed method decreases the packet drop, packet retransmission and end-to-end delay. Simulation results evidently show that the proposed scheme increases the network performance compared to other methods like semi-TCP, TCP-AP. COR describes the inelasticity of the collision which measures the performance of the network and also useful to make routing decision on the multi-hop environment. The benefits of CoRCiaR protocol are:
The traffic is distributed across all the 11 channels.
The reliability and connectivity are sustained in WMN.
Broadcasting and multicasting capabilities due to multiple channels.
97
Chapter 7
A Multi Route Rank Based Routing Protocol for Industrial Wireless Mesh Sensor Networks 7.1 INTRODUCTION A WSN comprises of tiny sensor nodes which are deployed in numerous environments, and networked together to establish coordination among the nodes to carry the detected events to the sink. The WSNs are widely used in various applications like military operations, frequency sensing, home automation, health monitoring, in marine engineering to sensor the underwater data, and industrial applications, etc. (Akyildiz et al., 2002; Romer and Mattern, 2004; Anastasi et al., 2009; Boukerche et al., 2007). The spatially distributed sensor nodes form a sensor network that can be used to monitor the environmental and physical conditions. The sensor nodes operate on minimal battery power and the lifetime of the nodes can be months to years. The Wireless Mesh Sensor Network (WMSN) (Poor, 2004) integrates the strengths of WMN and WSN, and it primarily brings in reliability, scalability and energy balancing. In industrial infrastructure, the traditional wired based communication mechanism was used, but due to its cost and the resource consumption, the trend has been shifted towards the wireless communication for data transfer. The legacy wired communication system in industries has been replaced by wireless sensor networks as the later offers great advantages like low cost of installation and maintenance (Gungor and Hancke, 2009; Akerberg et al., 2011), when deployed on a large scale. The Industrial Wireless Sensor Networks (IWSN) offers a variety of applications such as environmental monitoring, process monitoring, plant monitoring, and factory automation. The IWSN ensures the reliability and delivers the packets within a specified time among the nodes (Yoo et al., 2010). The existing routing protocols like AODV (Perkins and Royer, 1999), AOMDV (Marina and Das, 2001) and DSR (Johnson and Maltz, 1996) are not suitable for industrial environment considering its inability to withstand in hard environmental conditions, and electromagnetic interference (Agha et al., 2009).
98
Due to the harsh environmental conditions, the interference between the sensor nodes is high and that leads to transmission failure. The delay in process control and failure in reaching the destination within target time makes it unbearable and causes huge damage and financial loss in the industry. So, developing an approach with reliable and timely delivery of packets in the industrial environment became challenging. The existing reactive routing protocol fails to transmit the packets within a time frame due to dynamic conditions in industrial environment. The role of routing protocol is not only to find a path to the destination, it should also consider the delay, reliability and end-to-end throughput as these factors affects the productivity in the industry. In the hazardous industrial environment, sensor nodes suffer from huge radiations, high temperature, deep cold environment, ultraviolet radiations emitted by various industrial equipments, magnetic radiations by electromagnetic devices and so on. It may cause improper working of the sensor devices (or) may cause problems in communication among the nodes in the network, which would eventually result in loss of packets in the network. So, reliable communication is a worrisome and challenging problem due to varying channel conditions and node failures, which would lead to topology change and connectivity problem. The unreliability of the WSN reduces the throughput drastically. The challenges in IWMSN are:
The harsh environment causes the malfunction or failure of the sensor nodes.
Limited resources such as memory and battery power.
QoS constraints such as reliability and delay are difficult to achieve.
Data redundancy.
Packet error rate is high due to interference.
The reliable and timely transmission is not feasible using a single channel on industrial networks owing to the congestion and limited frequency. The current sensor nodes operate in multiple frequencies, and the proposed multi-channel is a solution to improve the reliability in industrial networks.
99
The multi-radio, multi-channel capability increases the capacity of the networks; and the contributions of this chapter are:
The forwarding node set is constructed by the route request phase and the rank of the node is calculated in multi-radio multi-channel scenario.
The actual forwarding candidate set is selected by prioritizing the link.
The SINR is a key factor which is used to compute the rank to quantify the interference on the link.
In this chapter, a Multi Route Rank based Routing (MR3) protocol is proposed, which facilitates to provide an alternate path in case of node failures and also identifies a path without interference for reliable delivery of information. This chapter is structured as follows: The Section 7.2 elaborates the existing reliable routing, opportunistic routing and energy efficient reliable routing in WSNs. The Section 7.3 presents the system model, construction of forwarding nodes set, actual forwarding candidates set and the rank of the node. The simulation setup, parameters, comparison and performance of MR3 protocol is analysed in Section 7.4. The chapter is concluded with Section 7.5.
7.2 RELIABLE ROUTING ALGORITHMS Zeng et al. (2010) had proposed an opportunistic routing approach, which is modeled as a linear programming to bind the channel to radio and to schedule the packet transmission. The Opportunistic Routing (OR) performs well in multi-channel multiradio environment, in comparison to the traditional routing. The OR utilizes less resource and produces better throughput. Hawbani et al. (2014) proposed a data routing approach, which divides the sensor nodes into distinct groups. In this approach, the intelligent adaptive scheme avoids flooding and ensures that each node receives only one copy of the message. The group leader selects the forwarding node to send the data along the base station. Marina and Das (2001) proposed an Ad-hoc on Demand Multipath Distance Vector Routing protocol (AOMDV) that extends the functions of AODV routing protocol. During link failures, this protocol efficiently finds the alternate route, and fast recovery is possible in dynamic networks. More than one route is computed through the 100
route discovery phase and it is primarily focused for high dynamic ad-hoc networks, where the link failures and path interruptions are most common. For every link failures, the route discovery is initiated and that consumes more time and resources. But AOMDV uses a single route request (RREQ) and finds multiple paths by accepting many RREQ packets. Johnson and Maltz (1996) proposed a Dynamic Source routing which is based on reactive method and it preserves the route cache in every node to maintain the path from source to the destination. During the data transmission, if there is any route failure, then the nodes should be updated with the path to the source node through the path establishment phase. Agha et al. (2009) proposed an OCARI technology which mainly developed for applications such as warships and power plants. This technology exploits the sensor nodes by using power aware routing and mesh topology. The energy is saved by keeping the network elements in sleep mode during the global cycle. Wang et al. (2010) presented a reliable routing protocol for IWSN to enhance the routing scheme through feedback and redundancy. The deterministic schedule is applied for energy saving and the data collection. The delay, routing metrics and the buffer size are not taken into the account. Heo et al. (2009) proposed an EARQ protocol for real time transmission which emphasizes on reliability and energy constraints. The demerit of this routing is that the overhead in exchanging control messages that are predominantly used for finding the global position of the nodes. Kim et al. (2011) proposed a reliable and energy efficient routing algorithm which considerably decreases the control packets and provides a reliable delivery of the packets. This method uses a single path to route the packets. Jamali et al. (2013) formulated routing as an optimization problem and it uses the BPSO to select a path which has the maximum energy and minimum hops. In WSNs, the nodes are deployed densely; the congestion occurs near the sink node, so a grid based approach (Shobana and Paramasivan, 2015) identifies the direction of all the nodes and then applies quorum methods to avoid congestion.
101
7.3 SYSTEM MODEL In this approach, a multi hop WMSN is considered with closely positioned nodes; each node in the network has more number of neighbours. When the nodes are deployed into wireless mesh sensor networks, they send hello handshake packets to the neighbouring nodes to find its density or the number of supporting nodes in its transmission range. On completion of this process, each node in the network would have identified the possible number of supporting nodes and the decision of the next hop selection is made based on the rank of the node. Fig 7.1 describes the overall architecture of multi route rank based routing protocols. This approach is a cross layer design between MAC and routing layers to increase the resilience to the dynamics of the link. The hop count, energy, number of supporting nodes and the noise ratio from the MAC layer is communicated to the routing layer to find the rank. The nodes are prioritized on the basis of the rank and multiple routes are made instead of a single route for a reliable packet transmission. Next hop node
Reliable Route Discovery
Selector
No of neighbors
Rank the nodes
Noise ratio
Hop count
Energy
Fig 7.1. Architecture of multi route rank based routing protocol
The channels are assigned based on the edge coloring algorithm to avoid adjacent channel interference, as the POC is used in multi-radio multi-channel IWMSN. The channel to interface binding is indicated in Fig 7.2, where the neighboring nodes are communicated if they are assigned with same frequencies.
102
Forwarding Nodes Set (FNS): It includes the nodes which are eligible to forward the packet and the eligibility is determined by the transmission range of the node, and also the node and its neighbours are tuned to the same channel. The route discovery process creates a set of possible paths from the origin to the destination and it is denoted as FNS = {F1, F2,.. Fi }, where i represents the maximum available path. F1={S, 1, 2, 3, 7, 8, 9, D} F2= {S, 1, 2, 7, 3, 9, D} F3= {S, 1, 2, 3, 9, D} F4= {S, 1, 2, 7,8,11, D} F5= {S, 3, 9, D} F6= {S, 4, 5, 6, 10, 12, D} F7={S, 1, 2, 3, 7, 8, 11, D} F8={S, 4, 5, 6, 9, D} Ch6 11 Ch11
8
Ch8
Ch6
D
Ch1
Ch11
7 Ch6 Ch11
Ch11 9
2 Ch6
Ch6
1
Ch11
Ch6 Ch11 12
3 10
Ch 1
Ch1
6 Ch1
Ch6 S
5 Ch11
Ch11 4
Ch1
Ch6
Fig 7.2. Multi-radio multi-channel in IWMSN
103
Ch1
Actual Forwarding Candidate Set (AFCS): It is a subset of FNS; these nodes will receive and send packets. It is constructed based on the rank assigned to the nodes. AFCS = {S, 3, 9, 10} 7.3.1 CONSTRUCTING FORWARDING NODE SET When a node is ready with data to the destination, the on-demand route discovery process is started if there is no recent path to the destination. Each sensor node in a wireless mesh sensor network must find a path to the destination before data transmission is initiated. This path must be reliable and cost effective to deliver the packets to the destination. In this process, the source creates a route request (RREQ) packets and broadcast it, if there is no existing path to the sink. As soon as the destination receives the RREQ, it will intern send a reply to RREQ along the same path or may be on different path. Route Request (RREQ) Process: When a node senses data on the environment, it forwards the data to the immediate neighbor, if it already has a path. Otherwise it floods RREQ packets. Each RREQ packet consists of the requester node identifier, destination to be reached, sequence number and the nodes visited so far. The intermediate node which receives RREQ finds its rank. Once the RREQ reaches the destination, Route Reply (RREP) message is exchanged by the destination. The hop count and the rank of each intermediate node are attached into the RREP packets. The source receives many RREP packets and using these, forward node set is constructed. The rank is calculated for every node to find the actual forwarding candidate nodes. It is computed by using the metrics hop count, SINR, node energy and the density. The node which is having less interference will be included in the actual forwarding candidate set. The rank considers the SINR and energy for better reliability and the rank is the key factor for path selection, with more energy, less interference and reliable. The node vj receives RREQ from the node vi, and the rank of a node vj is denoted by rij. The rij is defined as
104
rij
E(v j ) Hopcount t SINR ik SINR Kj 1 Density
E(v j )
(7.1)
E i (v j )
(7.2)
E r (v j )
Where the hop count is the distance between source and node vj, SINR is the Signal to Interference plus Noise Ratio, density is the number of common neighbors between node vi and vj, Er(vj) is the current residual energy in node vj , Ei(vj) is the initial energy in node vj and t is time slot. Ei(vj) is the maximum energy in joules in the deployed node. Er(vj) is the energy left in node vj. Let vk is a common node between vi and vj. The Signal to Interference plus Noise Ratio SINRik is greater than SINRij and also SINRkj is greater than SINRij. 1.75 1.62
11
1.9
8
Sink
7
1.6
9
2
1.7
D
1
3
1.4
12
1.11 10 6
S
1.64
5
Source
1.74
1.7 1.9
4 1.45
Fig 7.3 Forwarding RREQ packets along reliable path
The Fig 7.3 illustrates the rank calculation of the node. When the node receives RREQ packets, it calculates the rank and assumes that it is one of the forwarding 105
candidate nodes. For example, source S sends RREQ packets to neighboring nodes 1, 2, 3, 4, 5, and then all nodes compute their rank. Each node is maintaining a neighbor table, from that neighbor table 1, 2, 4 and 5 will be identified by 3 as mutual nodes. The rank of the node 3 is 1.11t according to equation 7.1. The priority is given to the route with more neighbors or supporters and the nodes with high energy level. Route reply packet is returned back to the source node by the sink node along the reverse path where the route request packet came from. At the end, the source builds the forwarding nodes set from the identified path. 7.3.2 ACTUAL FORWARDING CANDIDATE SET 1.75 1.62
11
1.9
8
Sink
7
1.6
9
2
1.7
D
1
3
1.11
1.4
{D}
{D, 9}
12 10
6 {D, 9, 3} S
5
Source
1.7
1.74
1.64
1.9 4 1.45
Fig 7.4 RREP packets forwarding
The Fig 7.4 is an example of route RREP propagation phase. For instance, 9 is the current forwarding node, then S (D, 9) will be attached in the RREP and the rank of the node 9. Similarly other nodes in the network also identify its neighboring nodes and attach them in RREP packets. Finally, the source gets multiple forwarding paths to ensure the reliability. The adjacent channel interference, interference from other sources, hop count and the energy of the node are evaluated at each node. The higher value of the rank indicates the stronger interference. The forwarding node set with minimum value as rank 106
is selected as the actual forwarding candidate set and the average rank is calculated for multiple received paths. The source node computes the average rank as follows:
Average Rank
r
FFNS
F
(7.3)
| F|
Where F is the forwarding nodes in set FNS and |F| is the number of nodes. The rank for the nodes r3= 1.11, r9= 1.4 So AFCS include (3, 9, D}, S-> 3->9->D is the actual optimal path to reach the sink. The rank of multiple paths and the average rank is indicated in Table 7.1. Table 7.1. Forwarding path and the average rank of the node
Path
Rank of the forwarding node
Average Rank
S, 1, 2, 3, 7, 8, 9, D
1.7, 1.6, 1.11, 1.9, 1.62, 1.4
1.16
S, 1, 2, 7, 3, 9, D
1.7, 1.6, 1.9, 1.11, 1.4
1.10
S, 1, 2, 3, 9, D
1.7, 1.6, 1.11, 1.4
0.96
S, 1, 2, 7, 8, 11, D
1.7, 1.6, 1.9, 1.62, 1.75
1.22
S, 3, 9, D
1.11, 1.4
0.62
S, 4, 5, 6, 10, 12, D
1.45, 1.9, 1.7, 1.64, 1.74
1.20
S, 1, 2, 3, 7, 8, 11, D
1.7, 1.6, 1.11, 1.9, 1.62, 1.75
1.21
S, 4, 5, 6, 9, D
1.45, 1.9, 1.7, 1.4
1.0
7.3.3 COOPERATIVE DATA FORWARDING Data generated at source is forwarded to a node with AFCS; we use cooperative forwarding method to avoid collision or the interference in the link. The node which is having the higher rank (less average rank) will start forwarding the packets and the rest will await their timer to be expired. The rank is interpreted as waiting time to forward the 107
data packet. If the node is assigned with higher rank or less waiting time, then it will get opportunity first to forward the data packets. Once the waiting time is expired, the node will start forwarding data to the upstream node and send ACK to the downstream node. The other lower priority nodes in the transmission range also hear ACK packets. If the node has not heard the ACK within a time window, the waiting timer expires and the next higher rank node will get its turn to forward the data packets. For example, from the node 1, 2, 3, 4 and 5, node 3 has higher rank or less waiting time, so node 3’s timer expires first compared to other nodes. Hence, the node 3 forward the data packets. Similarly, node 9 forwards the packet before other nodes in the next hop; hence the data traverses along the path Source->3->9->sink. The destination node sends ACK back to the source to suppress other nodes. The higher preference is given to the route with more neighbors or supporters and for the nodes with high energy level. Algorithm 1: Data forwarding at node vj Step 1. Void Data forwarding (packet p) Step 2. If node vj receives a packet p from node vi then Step 3.
Check if the received node (vj) is the destination then
Step 4.
Send ACK packet
Step 5. else if vj AFCS then Step 6.
call waiting timer(t)
Step 7.
Forward the data packets when waiting timer expires
Step 8. else vj AFCS then Step 9.
Waiting timer is expired but no ACK is received yet
Step 10.
Forward data packets
Step 11. end if
7.4 SIMULATION RESULTS The MR3 protocol is demonstrated in network simulator 2, and the comparison study carried out with other reactive routing protocols such as AODV-ETX and REPF (Hoe et al., 2004). The multi-radio, multi-channel patches are included in the Ns-2. 108
Density of the nodes is the maximum number of nodes installed in a certain area. The higher node density, then more reliable link and good connectivity. The simulation setup area is 200 m X 200 m square. The deployment of the nodes in any sensor environment is done randomly without any pre-specified metrics. The transmission range for each node is set as a radius of 50 m and the parameter T is set to 0.005 s. Not every route reply packet is acknowledged by the nodes in order to avoid the collision. Route reply is directly acknowledged by the receiver, not by the intermediate hop nodes. Performance of all three protocols are evaluated against diverse node densities and the results are shown in Fig 7.5, 7.6, 7.7 and 7.8, node densities are varied from 50 to 200. Table 7.2. Simulation parameters for MR3P
Parameters Platform Network Area Network Topologies Transmission Range Interference Range Frequency MAC Protocol Traffic Type Packet Size Maximum queue length Simulation Time Transport Type Data rate
Values Ns2 version 2.29 with multi radio multi-channel patch 1500 m X 800 m Chain topology and Random topology with multi radio 50 m 100 m 2.4 GHz IEEE 802.11 TCP 50 KB 50 100s TCP 1 Mbps
Fig 7.5 indicates the packet delivery ratio of different routing protocols in different node densities. The Multi Route Rank based Routing protocol achieved very high packet delivery ratio because of densely deployed nodes. MR3 protocol achieves 98% packet delivery ratio and the PDR increases with node density as shown in Fig 7.5. REPF achieves less PDR as the cooperation between the neighboring nodes is limited. Fig 7.6 describes the performance comparison of end-to-end delay against node density. Since AODV-ETX experiences more delay, comparatively, as the delay or time consumed for retransmitting the packet is much larger. 109
Fig 7.5. Density vs packet delivery ratio
Fig 7.6. Density vs. end-to-end delay
110
Fig 7.7. Throughput vs. density
Fig 7.8. Number of control messages vs. density
Fig 7.7 shows the throughput for MR3P, AODV-ETX and REPF. It is observed that the throughput of MR3P increases quickly along with the node density, though other protocols AODV-ETX and REPF also shown some increase with node density. With more number of forwarding nodes, the multi-channel increases the throughput in MR3P; the AODV-ETX and REPF uses single channel, so more competition for channel contention between the forwarding nodes. Fig 7.8 describes the comparison of control message cost against node densities. The control message cost of MR3P is slightly equal to REPF as retransmission of packets does not require control messages to be transmitted,
111
since the path is already identified at the beginning of the route establishment phase. The AODV-ETX has higher control message cost compared to other two routing protocols. The Table 7.3 shows that MR3P gives greater throughput over AODV-ETX & REPF. The end-to-end delay and number of control messages are decreased significantly. Table 7.3. Performance improvement of MR3P over AODV-ETX and REPF
ADOV-ETX
REPF
Throughput
+17%
71%
End-to-End delay
-35%
-7%
Number of control messages
-73%
-9%
7.5 CONCLUSION The proposed multi-route rank based routing protocol shown increase in the packet delivery ratio and throughput that clearly demonstrates more reliability on packet transmission over unreliable paths. The MR3P finds alternate path or link to transmit the data to the destination in a reliable way without re-establishing the connection. The adjacent channel interference, co channel interference, self-interference and interference from the external environment are reduced by finding the SINR on the link. The rank is introduced to select the better channel and path. The proposed MR3 protocol outperforms in multi-channel multi-radio environment and also surpasses other peer routing protocols like AODV-ETX and REPF.
112
Chapter 8
Conclusions and Future Enhancements In this research study, the main focus is set to explore the algorithmic techniques and cross layer optimization opportunities to improve the QoS in WMNs. The immense growth in wireless network technologies has attracted a number of users and various fields such as education, industries, government agencies and commercial shops, which led to increase in the user base, so do the network traffic. The basic issues in wireless networks such as capacity limitation, interference and scalability are addressed in this work. This work further delved into the issues of the routing and channel assignment in IEEE 802.11b/g WMN for orthogonal channels and POC. The interdependency between the channel assignment and the routing paves the way for cross layer design between MAC and network layers. The QoS is determined by both routing path and the link quality. The wireless networks are shared medium; many users compete to access the medium leads to interference. Therefore, we have addressed the cross layer routing and channel allocation strategies in WMNs for interference avoidance.
8.1 CONTRIBUTIONS The following are the major contributions of the research work:
In multi-channel, the adjacent channel interference is very severe, so there will be a significant amount of packet loss and that results in performance degradation. The channel assignment is presented as edge coloring problem, where complete spectrum is utilized and a new routing metric called SINR is presented, which computes the interference in every link and the information is fed to the routing algorithm for further processing. An approximation theorem postulating the edge coloring problem, which is otherwise an NP-hard problem, has been proved and utilized to good effect.
In MRMC-WMN, finding an optimal routing by satisfying the Quality of Service constraints is an ambitious task. The QoS guarantee is essential for 113
real time communications, but it is hard to achieve the QoS in wireless networks. An intelligent routing, using the hybrid PSO-GA is proposed to support QoS. This hybrid approach outperforms PSO and GA individually, and comparatively it takes less convergence time, by keeping away from the premature convergence.
Modification in the RTS/CTS method is proposed to upgrade the TCP performance. The CoRCiaR protocol is proposed to reroute the traffic in the congestion free path in WMN and the RTT in each hop is considered in making the routing decision. The CoRCiaR is completely novel in its class, exploiting collision theory in classical mechanics.
The Multi Route Rank based Routing (MR3) protocol is proposed to enhance the link dynamics for IWMSN and also provides interference free reliable packet delivery in harsh environments. The rank of a node is estimated based on density, hop count, the energy and SINR. The route discovery phase finds the rank value to forward the data packet in a reliable path.
The interference is modeled effectively by considering the distance between the nodes. The efficiency of proposed algorithms has been demonstrated through network simulations. It is concluded that, POC can significantly improve the overall performance of the network. The inefficient utilization of spectrum in IEEE 802.11b/g is eliminated in this proposed work and also supports more number of transmissions in dense networks. The main focus of this dissertation is to increase the number of devices accessing the Internet in IEEE 802.11b/g standard using all the available channels. The interference aware channel allocation and routing algorithms find QoS satisfied optimal path in WMN and provide reliable path in unreliable wireless networks.
8.2 FUTURE ENHANCEMENTS In this research work, some of the factors are considered for achieving the QoS and increasing the reliability. The possible future directions of the work are: 114
One of the important network parameter is transmission range of the node and it is considered as constant in this work. If the transmission power is adjusted dynamically, then there will be a significant increase in the network throughput.
The proposed research work has been carried out using the simulation to analyze the performance of the algorithm. The future directions of the research work are to find the efficiency of the algorithm in real life and conduct experimental research in the testbed.
The proposed channel allocation and routing algorithms are limited to IEEE 802.11b/g standard. The future direction is to continue this algorithm for other networks such as WiMAX (802.16) and IEEE 802.11a.
In future, the following cross layer framework is to be implemented:
Physical and transport cross layer scheme: It regulates the data transmission rate at the source in the transport layer by collecting information from the physical, to optimize congestion in the network.
Network and transport cross layer scheme: To aid the concept of traffic engineering.
115
PUBLICATIONS 1. Sarasvathi, V., and N. Ch. S. N. Iyengar (2012). Centralized Rank Based Channel Assignment for Multi-Radio Mutli-Channel Wireless Mesh Networks, Proc. of 2nd International Conference on Computer, Communication, Control and Information Technology (C3IT-2012), Procedia Technology, Hooghly, Vol. 4, pp. 182-186, West Bengal. 2. Sarasvathi, V., N. Ch. S. N. Iyengar and Snehanshu Saha (2014). An Efficient Interference Aware Partially Overlapping Channel Assignment and Routing in Wireless Mesh Networks, International Journal of Communication Networks and Information Security (IJCNIS), Vol. 6, No. 1, pp. 52-61. 3. Sarasvathi, V., N. Ch. S. N. Iyengar and Snehanshu Saha (2015). QoS Guaranteed Intelligent Routing using Hybrid PSO-GA in Wireless Mesh Networks, Journal of Cybernetics and Information Technologies, Vol. 15, No. 1, pp. 69-83. 4. Sarasvathi, V., Snehanshu Saha, N. Ch. S. N. Iyengar, Mahalaxmi Koti (2015). Coefficient of Restitution based Cross Layer Interference Aware Routing Protocol in Wireless Mesh Networks, International Journal of Communication Networks and Information Security (IJCNIS), Vol. 7, No. 3, pp. 177-186. 5. Sarasvathi, V., and N. Ch. S. N. Iyengar. A Multi Route Rank based Routing Protocol for Industrial Wireless Mesh Sensor networks. Journal of Cybernetics and Information Technologies. (Accepted).
REFERENCES Adya, A., P. Bahl, J. Padhye, A. Wolman and L. Zhou (2004). A Multi-Radio Unification Protocol for IEEE 802.11 Wireless Networks, Proc. IEEE First International Conference on Broadband Networks, San Jose, pp. 344-354, California. Åkerberg, J., M. Gidlund and M. Björkman (2011). Future Research Challenges in Wireless Sensor and Actuator Networks Targeting Industrial Automation, Proc. 2011 9th IEEE International Conference on Industrial Informatics (INDIN’11), Caparica, pp. 410415, Lisbon. Akyildiz, I. F. and X. Wang (2005). A Survey on Wireless Mesh Networks, IEEE Communications Magazine, Vol. 43, No. 9, pp. S23-S30. Akyildiz, I. F., W. Su, Y. Sankarasubramaniam and E. Cayirci (2002). Wireless Sensor Networks: A Survey, Elsevier Journal of Computer Networks, Vol. 38, No. 4, pp. 393422. AL Agha, K., M. H. Bertin, T. Dang, A. Guitton, P. Minet, T. Val and J. B. Viollet (2009). Which Wireless Technology for Industrial Wireless Sensor Networks? The Development of Ocari Technology, IEEE Transaction on Industrial. Electronics, Vol. 56, No. 10, pp. 4266–4278. Alicherry, M., R. Bhatia and L. E. Li (2006). Joint Channel Assignment and Routing for Throughput Optimization in Multi-Radio Wireless Mesh Networks, IEEE Journal on Selected Areas in Communications, Vol. 24, No. 11, pp. 1960-1971. Al-Jubari, A. M., M. Othman, B. M. Ali and N. A. W. A. Hamid (2011). TCP Performance in Multi-hop Wireless Ad hoc Networks: Challenges and Solution, EURASIP
Journal
on
Wireless
Communications
http://link.springer.com/article/ 10.1186%2F1687-1499-2011-198.
and
Networking,
Alzubir, A., K. Abubakar, A. Yousif and A. Abuobieda (2012). Article: State of The Art, Channel Assignment Multi-Radio Multi-Channel in Wireless Mesh Network, International Journal of Computer Applications, Vol. 37, No. 4, pp. 14-20. Anastasi, G., M. Conti, M. D. Francesco and A. Passarella (2009). Energy Conservation in Wireless Sensor Networks: A Survey, Elsevier Journal of Ad hoc Networks, Vol. 7, 2009, No. 3, pp. 537-568. Bahl, P., R. Chanda and J. Dunagan (2004). SSCH: Slotted Seeded Channel Hopping for Capacity Improvement in IEEE 802.11 Ad-hoc Wireless Networks, Proc. the ACM International Conference on Mobile Computing and Networking (MOBICOM ‘04), Philadelphia, pp. 216-230, Pennsylvania. Bahl, P., A. Adya, J. Padhye and A. Wolman (2004). Reconsidering Wireless Systems with Multiple Radios, ACM SIGCOMM Computer Communication Review, Vol. 34, No. 5, pp. 39-46. Balakrishnan, H., V. N. Padmanabhan, R. H. Katz (1999). Network Asymmetry: The Effects of Asymmetry on TCP Performance, Mobile Networks and Applications, Vol. 4, No. 3, pp. 219–241. Bhatia, R and M. Kodialam (2004). On Power Efficient Communication over Multi-hop Wireless Networks: Joint Routing, Scheduling and Power Control, Proc. Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies (INFOCOM 2004), Hong Kong, pp. 1457-1466, China. Bhorkar, A. A., T. Javidi and A. C. Snoeren (2015). Achieving Congestion Diversity in Multi-hop Wireless Mesh Networks, IEEE Transactions on Mobile Computing, Vol. 14, No. 3, pp. 473-483. Bokhari, F. and G. Zaruba (2012). On the Use of Smart Ants for Efficient Routing in Wireless Mesh Networks, International Journal of Wireless & Mobile Networks, Vol. 4, No. 2, pp. 117-134.
Boukerche, A., R. B. Araujo and L. Villas (2007). Optimal Route Selection for Highly Dynamic Wireless Sensor and Actor Networks Environment, Proc. the 10th ACM Symposium on Modeling, Analysis, and Simulation of Wireless and Mobile Systems, Chania, pp. 21-27, Greece. Brakmo, L. and L. Peterson (1995). TCP Vegas: End to End Congestion Avoidance on a Global Internet, IEEE Journal on Selected Areas in Communications, Vol. 13, No. 8, pp. 1465-1995. Brzezinski, A., G. Zussman and E. Modiano (2006). Enabling Distributed Throughput Maximization in Wireless Mesh Networks: A Partitioning Approach, Proc. 12th Annual International Conference on Mobile Computing and Networking (MOBICOM ’06), Los Angeles, pp. 26-37, California. Cai, Y., S. Jiang, Q. Guan and F. Yu (2013). Decoupling Congestion Control from TCP (semi-TCP) for Multi-hop Wireless Networks, EURASIP Journal on Wireless Communications and Networking (http://link.springer.com/article/10.1186%2F16871499-2013-149). Chandran, K., S. Raghunathan and S. R. Prakash (2001). A Feedback- Based Scheme for Improving TCP Performance in Ad hoc Wireless Networks, IEEE Personal Communications, Vol. 8, No. 1, pp. 34–39. Chaudhry, A. U., N. Ahmad and R. H. Hafez (2012). Improving Throughput and Fairness by Improved Channel Assignment using Topology Control Based on Power Control for Multi-Radio Multi-Channel Wireless Mesh Networks, EURASIP Journal on Wireless Communications and Networking, (http://link.springer.com/article /10.1186 %2F16871499-2012-155). Chen, C. C., H. Luo, E. Seo, N. H. Vaidya and X. Wang (2007). Rate - Adaptive Framing for Interfered Wireless Networks, Proc. 26th IEEE International Conference on Computer Communications (INFOCOM 2007), Alaska, pp. 1325-1333, Canada. Chen, T. W., J. T. Tsai and M. Gerla (1997). QoS Routing Performance in Multi-hop, Multimedia, Wireless Networks, Proc. 1997 IEEE 6th International Conference on
Universal Personal Communications Record, Conference Record, San Diego, Vol. 2, pp. 557-561, California. Chelliah, M., S. Sankaran, S. Prasad, N. Gopalan and B. Sevaselvan (2012), Routing for Wireless Mesh Networks with Multiple Constraints Using Fuzzy Logic, The International Arab Journal of Information Technology, Vol. 9, No. 1, pp.1-8. Cherif, M. A., M. K. Feraoun and S. B. Hacene (2013). Link Quality and MAC-Overhead Aware Predictive Preemptive Multipath Routing Protocol for Mobile Ad hoc Networks, International Journal of Communication Networks and Information Security (IJCNIS), Vol. 5, No. 3, pp. 210-218. Chiang, M (2004). To Layer or not to Layer: Balancing Transport and Physical Layers in Wireless Multi-hop Networks, Proc. Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies (INFOCOM 2004), Hong Kong, pp. 25252536, China. Chowdhury, K. R., M. Di Felice and L. Bononi (2013). XCHARM: A Routing Protocol for Multi-Channel Wireless Mesh Networks, Elsevier Computer Communication Journal, Vol. 36, No. 14, pp. 1485-1497. Crichigno, J., M. Y. Wu and W. Shu (2008). Protocols and Architectures for Channel Assignment in Wireless Mesh Networks, Elsevier Journal of Ad Hoc Networks, Vol. 6, No. 7, pp. 1051–1077. Cui, Y., W. Li and X. Cheng (2011). Partially Overlapping Channel Assignment Based on “Node Orthogonality” for 802.11 Wireless Networks, Proc. 2011 IEEE INFOCOM, Shanghai, pp. 361-365, China. Dai, L., Y. Xue, B. Chang, Y. Cao and Y. Cui (2008). Optimal Routing for Wireless Mesh Networks with Dynamic Traffic Demand, ACM Journal on Mobile Networks and Application, Vol. 13, No. 1-2, pp. 97-116. Das, A.K., H. M. Alazemi, R. Vijayakumar and S. Roy (2005). Optimization Models for Fixed Channel Assignment in Wireless Mesh Networks with Multiple Radios, 2005
Second Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks (IEEE SECON 2005), Santa Clara, pp. 463-474, California. De Couto, D. S. J., D. Aguayo, J. Bicket and R. Morris (2003). A High-Throughput Path Metric for Multi-Hop Wireless Routing, Proc. 9th Annual International Conference on Mobile Computing and Networking (MOBICOM ’03), ACM, San Diego, pp. 134-146, California. Dhananjay, A., H. Zhang, J. Li and L. Subramanian (2009). Practical, Distributed Channel Assignment and Routing in Dual-Radio Mesh Networks, Proc. ACM SIGCOMM 2009 Conference on Data Communication, Barcelona, pp. 99–110, Spain. Draves, R., J. Padhye and B. Zill (2004). Routing in Multi- Radio, Multi-Hop Wireless Mesh Networks, Proc. 10th Annual International Conference on Mobile Computing and Networking, ACM , Philadelphia, pp. 114-128, Pennsylvania. ElRakabawy, S. M. and C. Lindemann (2011). A Practical Adaptive Pacing Scheme for TCP in Multi Hop Wireless Networks, IEEE/ACM Transaction on Networking, Vol. 19, No. 4, pp. 975–988. El Masri, A., A. Sardouk, L. Khoukhi, A. Hafid and D. Gaiti (2014). NeighborhoodAware and Overhead-Free Congestion Control for IEEE 802.11 Wireless Mesh Networks, IEEE Transactions on Wireless Communications, Vol. 13, No. 3, pp. 58785892. Franklin, A. A., V. Bukkapatanam and C. S. R Murthy (2011). On the End-to-End Flow Allocation and Channel Assignment in Multi-Channel Multi-Radio Wireless Mesh Networks with Partially Overlapped Channels. Journal on Computer Communications, Vol. 34, No. 15, pp. 1858-1869. Gálvez, J. J., P. M. Ruiz and A. F. Skarmeta (2008). Distributed Algorithm for Gateway Load-Balancing in Wireless Mesh Networks, Proc. of 2008 1st IFIP Wireless Days Conference, Dubai, pp. 1-5, UAE.
Gungor, V. C and G. P. Hancke (2009). Industrial Wireless Sensor Networks: Challenges, Design Principles, and Technical Approaches, IEEE Transaction on Industrial Electronics, Vol. 56, No. 10, pp. 4258-4265. Hasan, M. I., S. Kumar (2015). Average Channel Capacity Evaluation for Selection Combining Diversity Schemes over Nakagami-0.5 fading Channels, International Journal of Communication Networks and Information Security (IJCNIS), Vol. 7, No. 2, pp. 69-79. Hawbani, A., X. Wang, Y. Xiong and S. Karmoshi (2014). Wireless Sensor Network Routing Based on Sensors Grouping, Wireless Sensor Network Scientific Research, Vol. 6, No. 1, pp. 8-17. Heo, J., J. Hong and Y. Cho (2009). EARQ: Energy Aware Routing for Real-Time and Reliable Communication in Wireless Industrial Sensor Networks, IEEE Transactions on Industrial Informatics, Vol. 5, No. 1, pp. 3-11. Huang, L and T. Lai (2002). On the Scalability of IEEE 802.11 Ad-hoc Networks, Proc. 3rd ACM International Symposium on Mobile Ad Hoc Networking and Computing (MOBIHOC ‘02), Lausanne, pp. 173–182, Switzerland. Holland, J. H. (1975). Adaptation in Natural and Artificial System: An Introduction with Application to Biology, Control and Artificial Intelligence. MIT Press Cambridge, MA, USA, pp. 92-109. Holland, G and N. Vaidya (1999). Link Failure and Congestion: Analysis of TCP Performance over Mobile Ad-hoc Networks, Proc. ACM Annual International Conference on Mobile Computing and Networking (MOBICOM), Seattle, pp. 219–230, Washington. IEEE 802.11 Standard Group Web Site. Available from: . Jacobson, V (1988). Congestion Avoidance and Control, Proc. SIGCOMM '88 Symposium Proceedings on Communications Architectures and Protocols, Seattle, pp. 314–329, USA.
Jain, K., J. Padhye, V. N. Padmanabhan and L. Qiu (2003). Impact of Interference on Multi-Hop Wireless Network Performance, Proc. 9th Annual International Conference on Mobile Computing and Networking (ACM MOBICOM ’03), San Diego, pp. 66-80, California. Jamali, S., L. Rezaei and S. J. Gudakahriz (2013). An Energy-Efficient Routing Protocol for MANETs: a Particle Swarm Optimization Approach, Journal of Applied Research and Technology, Vol. 11, No. 6, pp. 803-812. Jiang, H., W. Zhuang and X. Shen (2005). Cross-Layer Design for Resource Allocation in 3G Wireless Networks and Beyond, IEEE Communication Magazine, Vol. 43, No. 12, pp. 120–126. Johnson, D. B and D. A. Maltz (1996). Dynamic Source Routing in Ad hoc Wireless Networks. Mobile Computing, Springer US, pp. 153–181. Jou, T. S and D. E. Eastlake (2004). ESS MESH Network Study Group Meeting Minutes. Kaabi1, F., S. Ghannay, and F. Filali (2010). Channel Allocation and Routing in Wireless Mesh Networks: A Survey and Qualitative Comparison Between Schemes, International Journal of Wireless and Mobile Networks (IJWMN), Vol. 2, No. 2, pp. 132-150. Kennedy, J. and R. C. Eberhart (1995). Particle Swarm Optimization, Proc. IEEE International Conference on Neural Networks, Perth, Vol. 4, pp. 1942-1948, Australia. Khan, S. and J. Loo (2012). Cross Layer Secure and Resource Aware on Demand Routing Protocol for Hybrid Wireless Mesh Networks, Journal of Wireless Personal Communications, Vol. 62, pp. 201-214. Khurshid, A., M. H. Kabir and M. A. T. Prodhan (2011). An Improved TCP congestion Control Algorithm for Wireless Networks, Proc. 2011 IEEE Pacific Rim Conference on Communications Computers and Signal Processing (PacRim), Victoria, pp. 382-38, Canada.
Kim, W. and M. Gerla (2012). Cognitive Multicast with Partially Overlapped Channels in Vehicular Ad hoc Networks, Elsevier Journal on ad- hoc Networks, Vol. 11, pp. 20162025. Kim, M. K and H. P. Ngo (2011). A Reliable and Energy Efficient Routing Protocol in Industrial Wireless Sensor Networks Wireless Sensor Networks, Proc. 4th Annual International Conference on Advanced Technologies for Communications (ATC ’11), Da Nang, pp. 32-35, Vietnam. Kodialam, M. and T. Nandagopal (2005). Characterizing the Capacity Region in MultiRadio Multi-Channel Wireless Mesh Networks, Proc. 11th Annual International Conference on Mobile Computing and Networking (MOBICOM ’05), Cologne, pp. 73– 87, Germany. Kozat, U. C., I. Koutsopoulos and L. Tassiulas (2004). A Framework for Cross-Layer Design of Energy-Efficient Communication with QoS Provisioning in Multi-hop Wireless Networks, Proc. Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies (INFOCOM 2004), Hong Kong, pp. 1446-1456, China. Kyasanur, P., and N. H. Vaidya (2006). Routing and Link-Layer Protocols for MultiChannel Multi-Interface Ad hoc Wireless Networks, ACM SIGMOBILE Mobile Computing and Communications Review, Vol. 10, No. 1, pp. 31-43. Li, X and C. Xu (2009). Joint Channel Assignment and Routing in Real Time Wireless Mesh Network, Proc. IEEE Wireless Communications and Networking Conference, WCNC 2009, Budapest, pp. 1-6, Hungary. Lin, C. R and J. S. Liu (1999). QoS Routing in Ad hoc Wireless Networks, IEEE Journal on Selected Areas in Communications, Vol. 17, No. 8, pp. 1426-1438. Liu, Y., R. Venkatesan and C. Li (2010). Load-Aware Channel Assignment Exploiting Partially
Overlapping
Channels
for
Wireless
Mesh
Network,
Proc.
Telecommunications Conference (GLOBECOM 2010), Miami, pp. 1-5, FL.
Global
Liu, J., S. Singh (2001). ATCP: TCP for Mobile Ad hoc Networks, IEEE Journals on Selected Areas in Communications, Vol. 19, No.7, pp. 1300–1315. Marina, M. K and S. R. Das (2001). On-Demand Multipath Distance Vector Routing in Ad-hoc Networks, Proc. IEEE Ninth International Conference on Network Protocols, California, pp. 14–23, USA. Marina, M. K. and S. R. Das (2005). A Topology Control Approach for Utilizing Multiple Channels in Multi-Radio Wireless Mesh Networks, Proc. 2nd International Conference on Broadband Networks, Boston, pp.381 – 390, USA. Mascolo, S., C. Casetti, M. Gerla, M. Sanadidi and R. Wang (2001). TCP Westwood: Bandwidth Estimation for Enhanced Transport over Wireless Link, Proc. 7th Annual International Conference on Mobile Computing and Networking (MOBICOM ’01), Rome, pp. 287-297, Italy. Mishra, A., E. Rozner, S. Banerjee and W. Arbaugh (2005). Exploiting Partially Overlapping Channels in Wireless Networks: Turning a Peril into an Advantage, Proc. 5th ACM SIGCOMM Conference on Internet Measurement (IMC ’05), CA, pp. 1–6, USA. Mishra, A., V. Shrivastava, S. Banerjee and W. Arbaugh (2006). Partially Overlapped Channels not Considered Harmful, Proc. Joint International Conference on Measurement and Modeling of Computer Systems (ACM SIGMETRICS ’06 / Performance ’06), Saint Malo, pp. 63–74, France. Misra, J. and D. Gries (1992). A Constructive Proof of Vizing's Theorem. Journal of Information Processing Letters, Vol. 41, No. 3, pp. 131-133. Mockapetris, V (2006). Telephony's Next Act. IEEE Spectrum, Vol. 43, No. 4, pp. 1-5. Perkins, C. E and E. M. Royer (1999). Ad-hoc On-Demand Distance Vector Routing, Proc. Second IEEE Workshop on Mobile Computing Systems and Applications (WMCSA ’99), New Orleans, pp. 90-100, USA.
Poor, R (2003). Wireless Mesh Networks, In Sensors Magazine, Vol. 20, No. 2, pp. 3842. Poor, R (2004). Wireless Mesh Links Everyday Devices, Electronic Engineering Times. Prasad, R. and P. Kumar (2013). Joint Routing, Scheduling and Admission Control Protocol for WiMAX Networks, The International Arab Journal of Information Technology, Vol. 10, No. 1, pp. 85-94. Pries, R., D. Staehle, B. Staehle and P. Tran-Gia (2010). On Optimization of Wireless Mesh Networks Using Genetic Algorithms. International Journal on Advances in Internet Technology, Vol. 3, pp. 13-28. Rad, A. H. M. and V. W. S. Wong (2006). Joint Optimal Channel Assignment and Congestion Control for Multi-Channel Wireless Mesh Networks, IEEE International Conference on Communications (ICC ’06), Istanbul, pp. 1984-1989, Turkey. Ramachandran, K. N., E. M. Belding, K. C. Almeroth and M. M. Buddhikot (2006). Interference-Aware Channel Assignment in Multi-Radio Wireless Mesh Networks, Proc. 25th IEEE International Conference on Computer Communications (INFOCOM 2006), Barcelona, pp. 1–12, Spain. Raniwala, A and T. Chiueh (2005). Architecture and Algorithms for an IEEE 802.11Based Multi-Channel Wireless Mesh Network, Proc. 24th Annual Joint Conference of the IEEE Computer and Communications Societies, Florida, pp. 2223 – 2234, USA. Raniwala, A., K. Gopalan and T. C. Chiueh (2004). Centralized Channel Assignment and Routing Algorithms for Multi-Channel Wireless Mesh Networks, ACM SIGMOBILE Mobile Computing and Communications Review, Vol. 8, No. 2, pp. 50–65. Ren, J. and Z.Qiu (2009). Centralized Quasi-Static Channel Assignment for Multi-Radio Multi-Channel Wireless Mesh Networks, Scientific Research, Wireless Sensor Network, Vol. 1, No. 2, pp. 61-121.
Robinson, J., K. Papagiannaki, C. Diot, X. Guo and L. Krishnamurthy (2005). Experimenting with a Multi-Radio Mesh Networking Testbed, In WiNMee Workshop: 1st Workshop on Wireless Network Measurements, Seoul, Vol. 5, Korea. Römer, K., and F. Mattern (2004). The Design Space of Wireless Sensor Networks, IEEE Wireless Communications, Vol. 11, No. 6, pp. 54-61. Settles, M. and T. Soule (2005). Breeding Swarm: A GA-PSO Hybrid, Proc. 7th Annual Conference on Genetic and Evolutionary Computation (GECCO ’05), Washington, pp. 161-168, USA. Shobana, J and B. Paramasivan (2015). GCCP-NS: Grid based Congestion Control protocol with N-Sinks in a Wireless Sensor Network, International Journal of Communication Networks and Information Security (IJCNIS), Vol. 7, No. 2, pp. 99-105. Sinha, P., N. Venkitaraman, R. Sivakumar and V. Bharghavan (1999). WTCP: A Reliable Transport Protocol for Wireless Wide-Area Networks, Proc. 5th annual ACM/IEEE International Conference on Mobile Computing and Networking (MOBICOM '99), Seattle, pp. 231-241, Washington. Skalli, H., S. K. Das, L. Lenzini and M. Conti (2007). Channel Assignment Strategies for Multiradio Wireless Mesh Networks: Issues and Solutions, IEEE Communications Magazine, pp. 86-93. So, J. and N. H. Vaidya (2004). Multi-Channel Mac for Ad hoc Networks: Handling Multi-Channel Hidden Terminals using a Single Transceiver, Proc. 5th ACM International Symposium on Mobile Adhoc Networking and Computing (MOBIHOC ’04), Tokyo, pp. 222-233, Japan. Subramanian, A. P., H. Gupta, S. R. Das and J. Cao (2008). Minimum Interference Channel Assignment in Multi-Radio Wireless Mesh Networks, IEEE Transactions on Mobile Computing, Vol. 7, No. 12, pp. 1459–1473.
Tang, J., G. Xue, C. Chandler and W. Zhang (2006). Link Scheduling with Power Control for Throughput Enhancement in Multi-hop Wireless Networks, IEEE Transactions on Vehicular Technology (TVT), Vol. 5, No. 3, pp. 733-742. Tang, J., G. Xue, and W. Zhang (2005). Interference-Aware Topology Control and QoS Routing in Multichannel Wireless Mesh Networks, Proc. 6th ACM international symposium on Mobile Ad hoc Networking and Computing (ACM MOBIHOC ’05), Illinois pp. 68-77, USA. Wang, D., P. Lv, Y. Chen and M. Xu (2011). POCAM: Partially Overlapped Channel Assignment in Multi-Radio Multi-Channel Wireless Mesh Networks, 2011 11th IEEE International Symposium on Communications, Hangzhou, pp. 188-193, China. Wang, J., H. Zhai, W. Liu and Y. Fang (2004). Reliable and Efficient Packet Forwarding by Utilizing Path Diversity in Wireless Ad hoc Networks, 2004 IEEE Military Communications Conference, California, Vol. 1, pp. 258-264, USA. Wang, H., G. Ge, J. Chen and P. Wang (2010). A Reliable Routing Protocol Based on Deterministic Schedule for Wireless Industrial Networks, Proc. 3rd IEEE International Conference on Computer Science and Information Technology (ICCSIT ’10), Chengdu, Vol. 8, pp. 368–372, China. Wang, X and D. Perkins (2008). Cross-Layer Hop-by-Hop Congestion Control in Mobile Ad hoc Networks, Proc. IEEE Wireless Communication and Networking Conference, (WCNC), Las Vegas, pp. 2456–2461, USA. Wu, X., J. Liu and G. Chen (2006). Analysis of Bottleneck Delay and Throughput in Wireless Mesh Networks, Proc. 2006 IEEE International Conference on Mobile Ad-hoc and Sensor Systems, Vancouver, pp. 765–770, Canada. Yoo, S., P. K. Chong, D. Kim, Y. Doh, M. L. Pham, E. Choi, and J. Huh (2010). Guaranteeing Real-time Services for Industrial Wireless Sensor Networks with IEEE 802.15.4, IEEE Transactions on Industrial Electronics, Vol. 57, No. 11, pp. 3868–3876.
Zafar, H., D. Harle, I. Andonovic, L. Hasan and A. Khattak (2012). QoS-Aware Multipath Routing Scheme for Mobile Ad Hoc Network, International Journal of Communication Networks and Information Security (IJCNIS), Vol. 4, No. 1, pp. 1-10. Zeng, K., Z. Yang and W. Lou (2010). Opportunistic Routing in Multi-radio MultiChannel Multi-hop Wireless Networks, IEEE Transactions on Wireless Communications, Vol. 9, No. 11, pp. 3512-3521. Zhang, J., H. Wu, Q. Zhang and B. Li (2005). Joint Routing and Scheduling in MultiRadio Multi-Channel Multi-Hop Wireless Networks, Proc. 2nd International Conference on Broadband Networks, Boston, pp. 631-640, MA, USA. Zhao, Y., R. Sun and L. Xu (2004). An Ant Simulated Annealing Routing Algorithm for WMN, 2010 International Conference on Internet Technology and Application, IEEE Xplore, Wuhan, pp. 1-4, China. Zorzi, M., B. Zeidler, A. Anderson, B. Rao, J. Proakis, A. L. Swindlehurst, M. Jensen and S. Krishnamurthy (2006). Cross-Layer Issues in MAC protocol Design for MIMO Ad hoc Networks, IEEE Wireless Communications Magazine, Vol. 13, pp. 62-76. Zuo, Y., Z. Ling and Y. Yuan (2013). A Hybrid Multi-Path Routing Algorithm for Industrial Wireless Mesh Networks, EURASIP Journal on Wireless Communications and Networking (http://jwcn.eurasipjournals.com/content/2013/1/82).
