On the impact of Transport layer mechanisms on Routing Protocols in MANETS Anum Zaman
Sameer Qazi
Dept. of Electronics & Power Engg. PN Engineering College National University of Sciences & Technology (NUST) Islamabad, Pakistan
[email protected]
Dept. of Electronics & Power Engg. PN Engineering College National University of Sciences & Technology (NUST) Islamabad, Pakistan
[email protected]
Abstract—Wireless Ad hoc Networks have gained huge popularity due to the fact that the initial setup cost of decentralized networks with respect to centralized networks is negligible. Both Routing Layer and Transport Layer protocols have been proposed for Wireless Ad Hoc Networks to optimize the hop by hop path discovery and end to end data delivery. This paper discusses the performance analysis of wireless ad hoc routing protocols when used with different Transport layer mechanisms. Keywords—Mobile Ad hoc networks; Real and Non-real time traffic; Routing protocols; TCP Variants; Throughput; Normalized Routing Overhead
I.
INTRODUCTION
With the recent advancement in technology, the wireless networks are also gaining popularity and usage of wired technology has been limited as the number of internet users are increasing day by day. Wireless Networks can be classified in to Infrastructure and Peer to Peer Networks. Wireless LAN, Cellular Networks, WiMAX comes under the umbrella of Infrastructure Networks (Centralized Networks), where two mobile nodes can connect to each other via Base Station. On the other hand, Peer to Peer based networks (Decentralized Networks) do not require existing infrastructure and mobile nodes can connect to each other directly and are also called Wireless Ad hoc Networks. Ad hoc network which is infrastructure-less is a key to the evolution of wireless networks. Ad hoc network is formed by wireless devices connected to each other without the need of central control. To communicate with each other in a timely manner, routing protocols play an important role. New routes to the destination should be done with a minimum overhead and bandwidth consumption, thereby increasing the throughput of the wireless network. Demand is efficient access mechanism and the throughput it delivers to end users. Packet routing mechanism, number of intermediate nodes, distance between the nodes and speed of the nodes are the concerns when we are dealing with Ad hoc Networks. Routing protocols of mobile ad hoc network generally categorized as table driven and sourceinitiated [1]. Table-driven protocols maintain routing table at each node in the network regardless whether it is needed or not,
whereas source-initiated protocols maintain and update routing information only when needed. In this paper, we have briefly described one table driven protocol: Destination Sequenced Distance Vector (DSDV) and three source-initiated protocols: Ad hoc On-demand Distance Vector (AODV), Dynamic Source Routing (DSR) and Ad hoc On-demand Multipath Distance Vector (AOMDV). At the Transport layer, Transmission Control Protocol (TCP) and User Datagram Protocol (UDP) are two main and widely used protocols. Wireless medium is prone to rapid topological changes which results in packet loss. Slight modification in base algorithms of TCP i.e. Slow Start, Congestion Avoidance, Fast Retransmit and Fast Recovery led to the implementation of TCP variants, as basic TCP does not distinguish between losses due to mobility and enters into its slow start and congestion avoidance phase thereby reducing its throughput. To encounter this problem Freeze-TCP has been proposed by Goff et al. [2], which prevents from dropping its congestion window thereby increasing the throughput of the wireless network. This paper evaluates the performance of different of wireless ad hoc routing and transport layer mechanisms which have different objectives. Routing layer optimizes the hop by hop path discovery while Transport Layer ensures the end to end data delivery. Throughput and routing overhead has been used as evaluation parameters. Rest of the paper is organized as follows. Section II describes the related work; Sections III and IV describe the routing and transport protocols that we have analyzed. In Section V, simulation environment has been described and results are discussed in Section VI. Section VII concludes our paper. II.
RELATED WORK
Recent studies and work has been done on performance of routing in MANETs and transport protocols but mostly work has done considering only UDP traffic as presented in several studies [3, 4, 5]. On the other hand, comparison of different routing protocols for different terrain sizes at different mobility by using GLOMOSIM has been done by Panda et al. [6]. They
have done comparison of AODV and DSR. Comparison of TCP-Vegas and Reno has been done by Tabash et al. [7]. They have considered only two routing protocols AODV and DSDV. Dyer and Boppana [8], have done TCP analysis using TCP Reno by varying TCP connections and ad hoc routing protocols. Baig et al. [9] have proposed a prediction model for Freeze-TCP and showed its performance for on-board mobile networks. III.
AD HOC ROUTING PROTOCOLS OVERVIEW
In this section, we have given an overview of the ad hoc routing protocols which have been used for the performance analysis. A. Destination-Sequenced Distance Vector (DSDV) DSDV [10] is an adaption of Routing Information Protocol (RIP) and lies in the category of table-driven routing protocols. It overcomes its looping problem and distinguishes from stale routes entries in routing table by sequence number. The routing table of DSDV is maintained at each node. The sequence number is generated by destination node [10]. If the sequence number of update packet is same then smallest metric will be used and existing route will be discarded. B. Ad hoc On-Demand Distance Vector(AODV) In this routing protocol the routes are maintained only when required. The key characteristics of this protocol includes: path discovery packets are broadcast only when needed. Hello messages are sent to its neighbor nodes when local connectivity is needed [11]. Hello messages are also used for the detection of link. C. Ad hoc On-Demand Multipath Distance Vector(AOMDV) AOMDV is the extension of AODV [12]. AOMDV is an on-demand, multipath, loop free distance vector protocol which adopts the Route Discovery Mechanism of DSR and destination sequence number from DSDV. As based on distance vector concept, hop by hop routing approach is used by AOMDV. Route is learned on per need basis as needed by the traffic source. D. Dynamic Source Routing(DSR) Each node in DSR [13] maintains a Route Cache which stores source routes that it has learned. Once the packet is received by host, it reads the destination address first in Route Cache. If routing information is found, packet is forwarded to the next hop, if not; the forwarding host will use Route Discovery protocol to discover the source route to the destination. Due to the dynamic nature of the Wireless Ad hoc Nodes, Route Maintenance is used by DSR which detects a problem. IV.
TRANSPORT PROTOCOLS
In this section, we will give an overview of six connection oriented and one connectionless transport layer protocol which have been used in our simulations. A. TCP Tahoe TCP Tahoe is the simplest of all the TCP Variants. Tahoe starts to transmit in slow start phase [14] and window size is increased exponentially until it reaches the maximum safety
limit. On Packet Loss, throughput value goes to minimum and it re-enters in Slow Start Phase till the Slow Start threshold, then enters Congestion Avoidance phase. In this phase throughput value increases linearly. The problem with TCP Tahoe is that it sees every packet loss as a serious congestion and decreases the window size to minimum value. B. TCP Reno TCP Reno is another variant of TCP which checks Retransmission Timeout (RTO) and duplicates ACKs to classify congestion [15]. In case there is RTO, TCP Reno sees it as serious congestion and sets the congestion window size to minimum. If duplicate ACKs are received, congestion window size goes halved of the previous value as TCP Reno considers it as a non-serious congestion. TCP Reno’s fast recovery phase can only deal with single packet loss in congestion window and high degradation occurs when there is a multiple packet losses. C. TCP New-Reno TCP New-Reno [16] addressed the problem of TCP Reno and has the ability to detect multiple losses by staying in fast recovery state. On receiving triplicate acknowledgment it enters fast retransmit phase and remains in this phase till the acknowledgment of those packets that were outstanding when it enters in this phase. The benefit of remaining in fast retransmit phase is not to reduce the congestion window multiple times thereby giving better throughput. D. TCP SACK TCP Tahoe, Reno, New Reno all works on cumulative acknowledgements which may affect network performance. TCP SACK [17] uses selective acknowledgement and sends additional acknowledgements for packets which are not in sequence. This decreases the retransmission of data in event of single as well as multiple packet loss and helps to increase the throughput E. TCP Vegas TCP Vegas [18] detects congestion based on RTT Values and introduces the concept of Congestion Control. Congestion Detection highly depends on RTT Values. Vegas perform better when numbers of hops are increased as in MANETs. F. Freeze-TCP Freeze-TCP [2] is true end to end mechanism for mobile environment where the losses are mostly due to temporary disconnections (mobility, fading etc.). Upon predicting the losses by the receiver, it sends a zero window advertisement (ZWA). The sender upon receiving ZWA “freezes” it all data transmissions and enter into zero window probe (ZWP) mode, which prevents it from dropping its congestion window. It remains in the freeze mode until the hosts are reconnected. We have used the model developed by Baig et al. [9] for our simulations. We have briefly described the model which is based on Gilbert model [19] for wireless link and for detailed description about this model see [9]. In Markov model, there are two states: “up” and “down” and are error-free, means the losses are only due to link failure or when there is no communication between nodes which is mainly due to mobility. The link alternates between up and down state, which is exponentially distributed between tup and
tdown(tup=average uptime, tdown=average downtime).The probability of packets loss when in down state is denoted by pldown ,where l is link index. pldown is taken as 1 which means all the packets will be lost when the link is in down state and plup(probability of packets loss when is up state)