ChaMeLeon (CML), a hybrid and adaptive routing protocol designed to adapt to the rapid topological changes in extreme emergency MANETs (eMANETs) ...
A Hybrid Adaptive Routing protocol for Extreme Emergency Ad Hoc Communication Tipu Arvind Ramrekha and Christos Politis Wireless Multimedia & Networking (WMN) Research Group Kingston University London, United Kingdom Email: {a.ramrekha, c.politis}@kingston.ac.uk
Abstract—The autonomous nature of Mobile Ad hoc Networks (MANETs) makes them suitable for the support of extreme emergency rescuer IP communications for next generation networks. A major hurdle towards the deployment of emergency MANETs (eMANETs) is the design of a distributed routing protocol that can adapt to its highly dynamic topology where the number of nodes in the network frequently varies. This paper presents ChaMeLeon (CML), a hybrid and adaptive routing protocol designed to adapt to the rapid topological changes in extreme emergency MANETs (eMANETs) when operating within a predefined disaster area. CML adaptively changes its routing behavior according to the number of nodes in the network, so that it can provide a more efficient routing approach than purely proactive or reactive routing protocols for varying size networks. The protocol can operate in one of three routing phases which are the reactive phase (R-phase) for large networks, proactive phase (P-phase) for small networks and oscillation phase (O-phase) for phase transitions based on a network size threshold. This threshold is determined using simulation based performance statistics of the purely reactive or proactive routing protocols. We then present results to demonstrate that CML outperforms its routing counterparts over the simulated range of network sizes and particular critical area.
I. I NTRODUCTION Mobile Ad hoc Networks (MANETs) consists of a group of mobile and autonomous nodes that are directly interconnected to each other. These networks are independent of any communication infrastructures such as access points or base stations and rely entirely on node cooperation for relaying information from data sources to intended destinations. Such a self-organized network architecture makes MANETs suitable for extreme emergency communication frameworks as described in [1]. In disaster sites, where exiting communication infrastructures are often incapacitated or destroyed, rescuers would still be able to use extreme emergency MANETs (eMANETs) for multimedia communications for the purpose of rescue mission coordination and management. However, an eMANET specific routing protocol needs to be designed so that it operates in a distributed manner and is able to adapt to both logical and physical changes in the network state. In the context of eMANETs, factors that define the network state include node mobility levels, the node transmission radius, battery levels of nodes, traffic requirements, data load, wireless link quality and the network size. In extreme emergency cases, it is often the case that the network size will vary whenever more rescuers join or leave the disaster area or critical area (CA) according to the severity of the situation. In
addition, the rescuers would be equipped with lightweight mobile communication devices and battery exhaustion of communicating devices could stipulate another reason for changes in the size of the network. Thus, the eMANET routing protocol should provide efficient data communication in such dynamic topologies. The basic functionalities of any MANET routing protocol should include a route discovery mechanism, whereby routes are established between source and potential destinations or actual destinations, as well as a route maintenance mechanism which involves maintaining the validity of a route during data transmission. Currently, the two main routing approaches being discussed in literature are proactive and reactive. Nodes implementing proactive routing protocols such as the Optimized Link State Routing (OLSR) protocol [2], carry out route discoveries and route maintenances regularly in the shape of periodic exchanges of route information. Therefore, each node stores routes to all possible destinations at all times even if this information is not utilized. On the other hand, reactive protocols such as the Ad-hoc On Demand Vector (AODV) routing protocol [3], initiates route discovery on-demand whenever data has to transmitted to an actual destination. Finally, there is a hybrid routing approach where protocols combine the reactive and proactive routing behaviors of protocols in the aim of enhancing overall routing performance and making them more scalable. Some examples of such protocols include Sharp Hybrid Adaptive Routing Protocol (SHARP) [4], Zone Routing Protocol (ZRP) [5] and Landmark Ad Hoc Routing Protocol (LANMAR) [6]. It can be deduced from the above discussion that there is a tradeoff between proactive and reactive approaches. The proactive protocols pre-store route information and therefore decreases data delivery delay between source and destination due to routing layer functionalities while using high routing control overhead that scales with the number of nodes in the network. The overhead in this case is due to periodical route information dissemination which has no correlation with the timing or volume of data transmitted by the source node. Comparatively, the reactive routing approach increases the delay for end-to-end data packet delivery due to the latency of finding and maintaining routes on-demand. The routing overhead in the case of reactive protocols usually relates to the intervals of data transmissions as well the length of each transmission rather than being periodic. This results in a more scalable
978-1-4244-7116-4/10/$26.00 ©2010 IEEE
routing alternative due to reduced overhead. It is important to note here that assuming that our eMANET is based on the 802.11 technology, the CSMA/CA multiple wireless channel access protocol would be used. This implies that, for a given data load, higher routing overhead would introduce more delay towards packet deliveries. Hence, end-to-end delay for the scope of this paper consists of the sum of latencies introduced by both routing functionalities and lower layer transmission mechanisms. This paper describes and evaluates the ChaMeLeon (CML) routing protocol introduced in [7] and extended in [8]. This paper is based on the latter extension of the protocol. CML is designed to be used in eMANETs. It is a hybrid and adaptive routing protocol that operates within a pre-defined disaster area denoted as the Critical Area (CA). The main concept behind CML is its adaptability towards changes in the network state. In this paper, we intend to discuss the merits of CML as compared to other scalable protocols in literature. We also introduce the oscillation problem and provide possible solutions for such a problem. In a nutshell, the version of CML presented in this paper adapts its routing behavior according to changes in the network size within a pre-defined CA. The paper establishes a network size threshold (NST) on the basis of packet delivery delay and jitter performance metrics which are used as a measure of routing efficiency. The NST denotes the eMANET size beyond which AODV routing outperforms OLSR. Consequently and as justified in section IV, for small networks, CML routes data proactively using OLSR whereas for larger networks it utilizes the AODV routing protocol. These routing protocols are part of the CML proactive (P-phase) and reactive (r-phase) routing phases respectively, also referred to as the stable phases of routing. CML also introduces an oscillation phase (O-phase) which is an augmented version of the stable phases. Any transition between the stable phases has to occur via the Ophase to counteract effects associated with node oscillations as described in future sections. We also simulate the protocol against popular protocols such as AODV, OLSR and DYMO to compare their performances for small as well as large networks. II. R ELATED W ORK Currently, only purely reactive or proactive routing protocols are being discussed for standardization. The two most popular protocols are the proactive OLSR and the reactive AODV. OLSR [2] nodes periodically issues HELLO and Topology Control (TC) routing control packets for route discovery and route maintenance purposes. The Hello packets are sent to the one hop neighbor nodes for link quality information exchange while the TC packets are flooded network wide to exchange network topological information. The routing information received from these routing packets are then stored in the node routing tables and updated periodically as new information is received through routing packets. The particularity of OLSR is that it uses a flooding reduction mechanism whereby the TC packets are flooded only through
selected Multi-Point Relay (MPR) nodes for each node. The MPR node set selected by each node is computed by including the minimum number of bidirectional one hop neighbor nodes that covers all available two hop neighbors for that node. On the other hand, AODV[3] initiates route discovery on-demand when data transmission is required. It floods the network with a Route Request (RREQ) packet so that the destination node receives the RREQ. As a response, the destination generates a unicast Route Reply (RREP)to the user and thus establishes a source-destination (S-D) route. AODV nodes maintain established S-D routes in their routing table for a given timeout period after which route discovery has to be re-initiated for data transmission. As discussed above, while AODV and OLSR functionalities enable them to detect and use the availability of new router nodes in the network, they do not consider adapting to changes in the number of nodes within the network. For instance, the route discovery mechanism of OLSR will cause considerable and regular routing overhead for large networks. On the other hand, for smaller networks where OLSR routing might provide acceptable overhead for reduced data delivery delay, AODV will use less routing packets at the cost of added data delivery latency. While it is a good practice to minimize routing overhead to save battery power, the QoS of routing protocol in terms of delay is of high importance in the provisioning of multimedia communications as supported in [9]. There are hybrid, hierarchical as well as adaptive protocols, some that have been proposed in the context of large MANET sizes. The DYnamic MANET On-demand (DYMO) [10] Routing protcol is the successor of the AODV protocol and is currently a work in progress. It operates broadly in the same manner as its predecessor. Additionally, it introduces a Route Error (RERR) packet which is utilized to explicitly signal intermediate nodes towards the packet source and the source itself that a particular destination is invalid or missing. Upon receiving the RERR, the source node deletes the route through which that packet was obtained. Afterwards, if data needs to be sent to the same destination, a route discovery has to be re-initiated. Through this process, the authors intend to detect route breakages quickly so that packet losses and data retransmissions are minimized. The DYMO protocol introduces added routing overhead through the use of RERR packets. For larger networks, the route changes are more frequent and there might be a substantial amount of RERR packet overhead. The Zone Routing Protocol (ZRP) is described in [5] and was one of the first protocols with a hybird routing approach. ZRP nodes maintain a proactive zone radius covering a certain number of hop including neighborhood nodes. This proactive zone is varied according to the size of the network and the number of link failures. Outside of this proactive radius, the data is sent by reactively established routes. The main difference between ZRP and CML is that, the proactive ZRP route discovery involves using both a proactive and reactive mechanism in the network. ZRP solely focuses on minimizing routing overhead without considering delay and
jitter QoS metrics for multimedia traffic. Then, even in large networks, the intra zone routing mechanism still requires that node periodically sends route discovery and maintenance messages on top of any inter zone route discovery messages sent resulting in more overhead and delay than the CML protocol for such a scenario. LANMAR protocol [6] considers groups of MANET nodes as logical subnets of nodes. Within a subnet the node routes using a routing table where proactive information is stored. In addition, an elected node in each subnet acts as the landmark for that group. A second table is maintained for storing routes from source to landmarks. Therefore, any transmission intended within the subnet is sent proactively to the node. In case that the destination is located outside the source subnet, the data is sent towards the landmark node of the destination subnet which then routes it to the destination. LANMAR tries to segment the proactive message dissemination domain to reduce overhead in large networks. However, in the extreme emergency scenarios, group of nodes belonging to different emergency groups are expected to merge and partition. This will result in LANMAR nodes regularly electing new subnet landmarks as well as recalculating its tables frequently. Such a process will add more delay and overhead to the communication as compared to CML. Finally, SHARP is a hybrid routing protocol that uses AODV and OLSR. It describes various routing mechanisms of the protocol that addresses different network constraints and traffic requirements. One of the routing schemes try to address the issue of having a scalable routing mechanism for large networks. In this scheme, as opposed to shifting to a purely proactive or reactive routing as in CML, SHARP proposes to vary the OLSR proactive coverage radius of each node according to the size of the network. It then routes data using reactive AODV for destination nodes found outside this radius. However, although this scheme guarantees delay jitter performance improvements as compared to AODV, it cannot guarantee improvements in end-to-end packet delivery latency compared to both AODV routing and OLSR protocols. The Cluster Based Routing Protocol (CBRP) presented in [?] uses elected cluster heads to route messages in the network. All nodes proactively maintain route table information of nodes in the network. The neighbors of clusterheads cannot be voted as clusterheads themselves. Instead, one or more neighbor nodes could act as gateway nodes that are used for clusterhead-to-clusterhead communications. The routing process itself utilizes two routing tables which are the proactive table of two hop-neighbors generated by issuing HELLO messages and the cluster adjacency table (CAT), that stores information of other adjacent clusters. A source to destination data transmission is done as follows. RREQ packets are flooded to clusterheads which are responsible for broadcasting such packets. The gateway nodes receive such packets but only unicast it to the next cluterhead thus reducing packet overhead. Therefore, if a node receives a RREQ and finds that the destination is within the two hop neighborhood it sends it directly to that node. Nonetheless, as discussed
above, in large networks the concurrent operation of both reactive and proactive approaches can create a large amount of overhead. Also, the election of clusterheads might add substantial overhead in eMANETs where cluster members and node neighbors might change frequently. III. OVERVIEW OF CML CML operates in three distinct phases of operation which are O-phase, P-phase and R-phase. Each phase of operation comprises of an augmented version of a routing protocol so that it interacts with the CML Adaptive Module. In the event that the routing protocol receives a routing control packet in the stable phase of operation, it contacts the ”Monitor” function of the Adaptive Module by passing the current routing phase information. The latter function checks if the network size threshold is exceeded and calls the ”Adapt” function in that case. Otherwise the normal routing processing defined in that operation phase is resumed. The ”Adapt” function initiates the O-phase of operation passing on information about the current routing phase and the nature of the method whereby a threshold breach was detected. In the O-phase of operation, the routing processes as defined by the current operation phase continues while the O-phase uses routing specific CML mechanisms to confirm whether the size of the network actually necessitates a change in phase. A change in phase is only allowed after a pre-defined time interval referred to as the oscillation interval (Osc Interval) which is reset each time a phase shift occurs. The O-phase is used to both confirm the actual size of the network and to prevent node oscillation as defined below. Therefore, any change from one stable phase to the other, called a ”phase shift”, has to be made through the O-phase. If it is confirmed that the network threshold has been exceeded, the O-phase allows a phase shift. A. Routing Phases and the Adaptive Module The various processes involved in the “Adaptive Module” and O-phase varies according to the stable phase that was responsible for initiating the call. In P-phase operation, which is also the default phase of operation, the routing is carried out using OLSR. When a TC message is received and the routing table is updated, the Monitor function of the “Adaptive Module” counts the number of nodes, “N”, in the network which is the summation of destination nodes from the table. The value of N is compared with the network size threshold, “NST”. If the value of N is greater than NST, the“Adapt” function is called with current phase operation and the context through which it is being initiated, otherwise the normal routing processes of the P-phase is resumed. The Adapt function, called in this case using the context of “Nsize” (for direct determination of N being greater than NST), then initializes the O-phase. In the O-phase operation, the O-phase validity time, “Osc Interval” of the oscillation timer is first checked and if this is not expired, the stable phase operation is resumed. OLSR algorithm will continue to handle the routing process and the number of nodes will be checked for a time period of 2*TC Intervals (the term TC Interval is described
in [2]). If at least one more count of the number of nodes is found to be greater than NST, the O-phase switches to Rphase operation and resets the oscillation timer to the value of “Osc Interval”. It also generates and floods a CML Change Phase (CP) Packet which signals neighborhood nodes of such a phase shift. In the R-phase operation, AODV is used as the routing protocol. When a RREP packet is received at a node, the “Monitor” function of the “Adaptive Module” is called, with current phase information, in order to obtain a value for “N”. In this case, assuming that the nodes are evenly distributed throughout the network where it can be deduced that N approximates to the value from the function F (HC 2 ), the value of N is first estimated using the maximum value of the hop count (HC) from a source node to a destination.Also, the RREP packets contain a value of HC that can be used towards that end. If the estimated value of N is less than or equal to NST, the“Adapt” function is called with current phase as the “R-phase” and the context of ”Nsize”, otherwise the normal routing processes of the R-phase is resumed. The “Adapt” function instantiates the O-phase operation. The O-phase process includes first checking the validity of the oscillation timer. If the timer is still valid, the stable phase of operation is resumed. Otherwise, the AODV routing protocol will route packets normally and sends data through the previously established route. After data transmission, a CML HC-Request (HCREQ) packet is generated and flooded in the network to probe for the network HC max (as opposed to the previously received destination HC). The HCREQ will have a TTL √ calculated using the inverse of the function above i.e F ( N ST ) so that the flooding is limited and the minimum number of hops indicating that the value of “N” is greater than NST, can be covered.Also, each node receiving the HCREQ with a TTL greater than zero, generates its own echo-HCREQ and floods it in the MANET. In case an HCREQ or echo-HCREQ with a TTL value of zero is received, a node has to generate a corresponding HC Reply (HCREP) and unicast it to the source node. If a node receives the HCREP for an echo-HCREQ it generated, it has to send a gratuitous HCREP to the source node. In this manner, CML makes sure that the network HC has been inspected from nodes located at different positions within the eMANET. However, if the source node does not receive a HCREP within 4*NET TRAVERSAL TIME (described in [3]) of sending an HCREQ it determines that the current value of “N” is less than “NST”. The mode of operation is therefore changed to P-phase and the CML CP packet is flooded in the network. If a node receives a CML CP packet in any operating stable phase, it calls the “Adapt function” of the “Adaptive Module” directly where the context of network size change is set as “CP” (for indirect determination that N is unsuitable for current mode of operation made by neighbor nodes). The initialized O-phase only checks for an invalid oscillation timer to allow a phase shift in the case that the context is “CP”. It must also reset the oscillation timer to the value of “Osc Interval” and flood the CP packet back in the network after decrementing its TTL count. However, if the oscillation
timer is found to be still valid when first checked, current stable phase of operation is resumed and no phase shift is allowed. It is also important to note that CML can prevent routing loops by using a table to store the sequence number and packet source information of CML packets as explained in both [3] and [2]. A summary of the CML protocol operation is shown in Figure 1.
P-phase •Receive Roung Packets 1a
•Contact Adapve Module
2a
Adapve Module
R-phase
Monitor (Phase flag)
•Receive Roung Packets
•If Network Size Threshold (NST) exceeded, Call Adapt funcon •Else return to stable phase (2(a) or 2(b))
1b 2b
•Contact Adapve Module
Adapt (Phase flag, context, call)
•Receive Data Packets 5 5a
•Receive Data Packets
•Inialise O-phase
3
4
5b
O-phase •Connue current stable phase of operaon •Check Oscillaon Timer : If valid go back to stable phase (4)->(2(a)or 2(b)) •Check Ch k (context, ( phase h flag fl and d call) ll) values: l If (N (Nsize, i p-phase, h monitor), i ) check node count validity; If (Nsize, r-phase, monitor), check hop count validity ; If (Nsize, *, CML_CP), check CP packet validity ; •If validity check fails, go back to stable phase (4)->(2(a)or 2(b))) •Switch to R-phase or P-phase if checks are successful (5(a)or 5(b)) •Send Send CP packets to alert neighbours of phase change. change •Reset Oscillaon Timer.
Fig. 1.
Overview of the CML protocol operations.
B. Emerging Challenges and CML Solution The two main challenges that arise when designing such a hybrid and threshold-based adaptive protocol as CML are the way node oscillations can be tackled and determination of the size threshold value. Firstly, oscillation occurs when nodes join and leave the network repeatedly so that the value of “N” fluctuates beyond and falls below the NST value on a sequential basis. This is usually the case in emergency scenarios and would cause performance degradation in CML due to frequent phase shifts as well as CML CP packet flooding. To tackle this issue, we characterize oscillation by the number of nodes that oscillates and the frequency of oscillation. We propose a two-fold solution whereby appropriate different NST values are used depending on the current stable phase of operation in the network that is we use an Upper-NST (U-NST) for Pphase and Lower-NST (L-NST).As a result, if for a particular network there is a tendency for α nodes to oscillate in the network U -N ST = N ST +α and L-N ST = N ST −α. These phase specific thresholds will restrain the negative effects of group oscillations on CML. On the other hand, if nodes are
IV. E VALUATION
End-to-end d-to-end -end d Delay lay in seconds econds nds
The evaluation section has two parts. Firstly, the performance of OLSR and AODV are compared to determine an NST value for CML. Then, the CML protocol is simulated to compare its performance against OLSR, AODV and DYMO. The performance results of the Dynamic State Routing (DSR) protocol described in [11] is also included purely for the sake of comparison. The simulation scenario is described next. The simulation area is 1000m*1000m, the average network node speed 1.5 m/s and the value of average node pause time of 10 seconds using the emergency HUMO Mobility Model as presented in [12]. We use a CBR traffic of 64 kbps to simulate the use of voice data transmission over the network with 10 CBR connections. The number of nodes for which the simulation scenario was run are 5, 10, 15, 20, 25, 30, 35, 40, 45 and 50 nodes. The Figures 2-5 show the simulation results from the above scenarios. 3 2.5 2 1.5 1 0.5 0
CML DSR OLSR AODV DYMO
5
10
15
20
25
30
35
40
45
50
0.155376 2.77068 0.152184 2.70938 2 70938 2.5673 2 5673
0.027087 0.139621 0.021338 0.166203 0 166203 0.1452 0 1452
0.308904 0.425463 0.312801 0.305771 0 305771 0.313031 0 313031
0.357967 0.505816 0.358851 0.357098 0 357098 0.318904 0 318904
0.336344 0.523277 0.410214 0.334221 0 334221 0.324716 0 324716
0.372783 0.577201 0.56449 0.372783 0 372783 0.35569 0 35569
0.197011 0.387228 0.392517 0.207011 0 207011 0.200765 0 200765
0.26843 0.399775 0.418757 0.267446 0 267446 0.250735 0 250735
0.269315 0.769935 0.530194 0.269315 0 269315 0.300734 0 300734
0.228047 0.647567 0.440852 0.228047 0 228047 0.228392 0 228392
Network Size in nodes
Fig. 2.
end-to-end delay for each routing protocol against network size.
Cumulative mulative tive Delayy Jitter ter
14 12 10 8 6 4 2 0
5
10
15
CML 0.0120213 0.142829 0.694719 DSR 0.0190892 0 0190892 1.65616 1 65616 2.32066 2 32066 OLSR 0.0120213 0.123546 1.1079 AODV 0.0162045 0 0162045 0.18943 0 18943 0.749197 0 749197 DYMO 0.01747 0.1935293 0.7802978
20
25
30
1.51544 2.25423 3.08183 3 23729 4.53008 4 53008 5.24003 5 24003 3.23729 2.20676 3.05369 5.62891 1 57228 2.30899 2 30899 3 13659 1.57228 3.13659 1.606507 2.104733 3.048034
35 4.47838 6 51881 6.51881 7.61124 4 53314 4.53314 4.61001
40
45
50
5.58247 6.27976 7.63934 8 77179 9.57226 9 57226 10.5361 10 5361 8.77179 10.3896 11.0653 12.1154 5 62832 6.32561 6 32561 7.68519 7 68519 5.62832 5.553038 6.387445 7.737046
Network Size in nodes
Cumulave Average End-to-End End Packet Delivery in Secondss
Fig. 3. size.
Cumulative packet jitter for each routing protocol against network
8 7 6 5 4 3 2 1 0 5
10
15
20
25
30
CML
0.155376
0.0270867
0.308904
0.357967
0.336344
0.372783
0.197011
0.26843
0.269315
0.228047
DSR
2.77068
2.910301
3.335764
3.84158
4.364857
4.942058
5.329286
35
5.729061
40
6.498996
45
7.146563
50
3.602198
OLSR
0.152184
0.173522
0.486323
0.845174
1.255388
1.819878
2.212395
2.631152
3.161346
AODV
2.70938
2.875583
3.181354
3.538452
3.872673
4.245456
4.452467
4.719913
4.989228
5.217275
DYMO
2.5673
2.7125
3.025531
3.344435
3.6691507
4.0248407
4.2256057
4.4763407
4.7770747
5.0054662
Network Size in nodes
Fig. 4. Average cumulative packet end-to-end delay for each routing protocol against network size.
Total Roung Load in bytes
expected to oscillate at particular time periods of τ seconds, an oscillation timer is utilized as described above using a timer validity time of Osc Interval = τ . Thus no phase shift is allowed due to periodic oscillations. In addition, during the O-phase, a node checks for more instances of the value of “N” and confirms that the first monitored value of “N” was not a result of oscillation. CML also needs to determine a value for NST which will act as a balance point in terms of the network size wherein OLSR will outperform AODV. For this purpose we have decided to use the data packet delivery delay and jitter metrics as a measure of protocol efficiency so that the protocol efficiency will be optimized for multimedia transmission in eMANETS. Firstly, the NST value could be determined using a mathematical model for the routing protocol operations and the network state. In this paper, we choose to establish the NST by using simulation statistics. To do so, the OLSR and AODV routing protocols are simulated for a given scenario where the variable parameter is the number of participating eMANET nodes. The performance of these protocols in terms of data delivery delay and jitter against varying number of nodes in the network, are then illustrated using bar graphs. The NST value is set to the network size where one of the protocols out performs the other based on both delay and jitter measurements. A more detailed description of the simulation setup and the quantitative determination of NST is described in the section IV.
1000000 900000 800000 700000 600000 500000 400000 300000 200000 100000 0 5
10
15
20
25
30
35
40
45
50
CML
16963
36198
112980
124385
262034
254806
732424
585925
574462
938816
DSR
2503
47284
128516
151718
376889
246228
475150
379475
892444
948931
92512
138038
183972
227918
258762
352557
390646
OLSR
16963
35633.5
64284
AODV
1915
23383
106005
121686
261176
251693
729036
583632
570117
934245
DYMO
16095
32052
155095
183414.5
300275
264532
796846
592865
632625
945694
Network Size in Nodes
Fig. 5.
Total routing load for each routing protocol against network size.
Firstly, it can be seen from Figure 2 that the end-to-end delay performance of OLSR is better than AODV for network sizes of less than or equal to 10 nodes and that AODV performs better for networks greater than 10 nodes. Additionally, in Figure 3 the same situation can be observed where the endto-end data delivery jitter due to OLSR operation is better than AODV routing for networks of size less than or equal to 10 nodes whereas AODV outperforms OLSR for bigger network sizes. We therefore choose the value of N = 10 as the NST for CML. The values of LNST, UNST and Osc Interval are subject to network situations and should be set according to the expected behavior of rescuers in an extreme emergency scenario. Then, when CML is evaluated against AODV and OLSR for the above described scenario, it can be observed in Figure 2 and Figure 3 that CML outperforms AODV for small sized networks of up to N nodes and outperforms OLSR for bigger networks of up to 50 nodes in terms of both data end-to-end delivery delay and jitter. Also it can be observed that the delay
performance for nodes of less than or equal to 10 nodes is slightly higher than OLSR. Then, for bigger networks, CML performs slightly worse than AODV because of transmission and processing delay of CML CP packets. However, Figure 2 illustrates that CML will provide a better performance than both AODV and OLSR for varying size networks and that it is a better alternative than both these protocols for such growing or shrinking networks. The Figure 5 shows the cost of using the protocols in terms of the routing control packets that they utilize to function. It is clearly noticeable that CML uses approximately the same amount of overhead as OLSR for small networks and as AODV for networks larger than 10 nodes. Therefore the added routing cost due to CML CP packets can be regarded as negligible as compared to the improvement it provides in terms of delay and jitter. It should also be noted that DSR has the worst delay and jitter performances as compared to the other protocols although it uses the least routing overhead. Therefore, using delay and jitter performance metrics, we can deduce that the proactive OLSR routing approach is better for small networks whereas for larger networks the reactive AODV protocol will provide a better routing alternative in eMANET scenarios such as the one described above. CML outperforms the protocols because it adaptively uses proactive or reactive approaches which best suits the size of the network. Additionally, the performance of DYMO is similar to that of AODV based on the metrics and over the range of network sizes considered. From the above figures, it can be noticed that DYMO only uses slightly higher routing overhead than AODV while also being only slightly outperformed delay and jitter wise by AODV. However, DYMO does propose a faster approach towards route failure detection and reparation than AODV. This can in turn help in the reduction of packet loss and data retransmission overhead. V. C ONCLUSION The eMANET features nodes that regularly joins and leaves the network based on emergency situations. Therefore, there is a need for a routing protocol that can perform adequately in this context. We have also established for a given simulation scenario and based on end-to-end delay and jitter performance metrics that the proactive approach of OLSR better suits MANETs with a network size of less than or equal to 10 nodes whereas the reactive AODV protocol is better for larger networks. We have proposed the CML routing protocol that can adapt to the varying network size using OLSR in small networks and AODV for larger networks such that it improves the overall efficiency of these approaches. It can be concluded that an adaptive and hybrid protocol such as CML can be used as a better alternative than OLSR and AODV for supporting multimedia communications in eMANETs. Future work includes comparing the performance of AODV against DYMO in more details. Based on the current simulation results, DYMO does provide an interesting mechanism to detect and repair route failures for a slightly higher cost than
AODV. Therefore, this mechanism could be usefully included in the CML protocol. VI. ACKNOWLEDGEMENTS The authors wish to acknowledge the support of the ICT European Research Programme and all the partners in PEACE: PDM&FC, Instituto de Telecomunicaes, FhG Fokus, University of Patras, Thales, Telefonica, CeBit. R EFERENCES [1] Fokus et al., “Functional Models and plan scenarios,” ICT FP7 PEACE project, ICT-SEC-2007.1.7, Deliverable D2.3, Nov. 2009. [2] T. Clausen and P. Jacquet, “RFC3626: Optimized Link State Routing Protocol (OLSR),” RFC Editor United States, 2003. [3] C. Perkins, E. Belding-Royer, S. Das et al., “RFC3561: Ad hoc ondemand distance vector (AODV) routing,” RFC Editor United States, 2003. [4] V. Ramasubramanian, Z. Haas, and E. Sirer, “SHARP: A hybrid adaptive routing protocol for mobile ad hoc networks,” in Proceedings of the 4th ACM international symposium on Mobile ad hoc networking & computing. ACM New York, NY, USA, 2003, pp. 303–314. [5] Z. Haas, M. Pearlman, and P. Samar, “The zone routing protocol (ZRP) for ad hoc networks,” draft-ietf-manet-zone-zrp-02.txt (work in progress), 1999. [6] G. Pei, M. Gerla, and X. Hong, “LANMAR: landmark routing for large scale wireless ad hoc networks with group mobility,” in Proceedings of the 1st ACM international symposium on Mobile ad hoc networking & computing. IEEE Press, 2000, p. 18. [7] T. A. Ramrekha and C. Politis, “An adaptive qos routing solution for manet based multimedia communications in emergency cases,” in Mobile Lightweight Wireless Systems, F. Granelli, C. Skianis, P. Chatzimisios, Y. Xiao, and S. Redana, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2009, vol. 13, ch. 8, pp. 74–84. [Online]. Available: http://dx.doi.org/10.1007/978-3-642-03819-8 8 [8] T. A. Ramrekha et al., “ChaMeLeon (CML): A hybrid and adaptive routing protocol for Emergency Situations,” IETF Internet Draft, http://tools.ietf.org/html/draft-ramrekha-manet-cml-00.txt (work in progress), March 2010. [9] P. Mohapatra, J. Li, and C. Gui, “QoS in mobile ad hoc networks,” IEEE Wireless Communications, vol. 10, no. 3, pp. 44–53, 2003. [10] I. Chakeres and C. Perkins, “Dynamic manet on-demand (dymo) routing,” draft-ietf-manet-dymo-19 (work in progress), 2010. [11] J. Broch, D. Johnson, and D. Maltz, “The dynamic source routing protocol for mobile ad hoc networks,” Internet-Draft, draft-ietf-manetdsr-00.txt(work in progress), Tech. Rep., 1998. [12] C. Papageorgiou, K. Birkos, T. Dagiuklas, and S. Kotsopoulos, “An obstacle-aware human mobility model for ad hoc networks,” in Modeling, Analysis & Simulation of Computer and Telecommunication Systems, 2009. MASCOTS ’09. IEEE International Symposium on, Sept. 2009, pp. 1–9.