AODV enhancements in a realistic VANET context
Jonathan Ledy, AnneMarie Poussard, Rodolphe Vauzelle Laboratoire XLIMSIC UMR CNRS 6172 Université de Poitiers, France [ledy, poussard, vauzelle]@sic.univpoitiers.fr Abstract—Ad hoc On demand Distance Vector (AODV) is a commonly used routing protocol for Vehicular Ad hoc NETworks (VANET). This paper presents and analyzes several AODV enhancements propositions dedicated to the VANET context. Communication protocol tuning can yield significant gains in energy efficiency, resource requirement, and overall network performance, all of which is of particular importance in VANETs. Alternatively, multipath routing allows the establishment of multiple paths between a pair of source and destination nodes in mobile ad hoc networks. It has recently received more and more attention and is typically proposed to increase the reliability of data transmission. This paper shows how AODV tuning and mutipath routing behave under realistic VANET simulations. KeywordsVANET; Routing protocol; AODV; NS2
I.
INTRODUCTION
An ad hoc wireless network consists of mobile nodes that communicate with each other without infrastructure. Vehicular Ad hoc NETwork (VANET) is an emerging new type of such networks in which vehicles constitute mobile communicating nodes. This kind of network supports both vehicletovehicle (V2V) and vehicletoinfrastructure (V2I) communications by providing multihop and selforganized networks. In this work, we focus on V2V communications. The main technical challenges for those communications are the strong mobility of the nodes, the highly dynamic change of the topology and the variability in node density which result in a high versatility of the wireless communication links. Because there are no dedicated access points nor routers in a VANET and due to the high mobility, the routing algorithms are one of the important subjects of adhoc networks. Therefore, many routing protocols for adhoc networks have been proposed. One of the most popular is Ad hoc On demand Distance Vector [1] (AODV). It is a singlepath, reactive routing protocol. Route discovery uses a Route REQuest (RREQ) – Route REPly (RREP) cycle. Ondemand approach is not without problems. Since routes are computed only on demand, route discovery latency can add to the endtoend delay, until a previously computed route is available. Buffering of data packets during the route discovery process
Benoît Hilt, Hervé Boeglen Laboratoire MIPSGRTC EA2332 Université de Haute Alsace, France [herve.boeglen, benoit.hilt]@uha.fr
can also contribute to packet losses due to buffer overflow. This is why there exist many works that propose new algorithms derived from the standard AODV. For example, the SAODV routing protocol proposed in [4] improves AODV security. VAODV proposed in [5] uses crosslayer metric in order to improve QoS, while the authors of [6] propose DFIAAODV, an interference aware routing scheme for the same purpose. Perkins et al., the creators of the AODV protocol, discussed the reasons for applying hello messages with AODV and presented some drawbacks for using these messages. Yet, some works [2] focus on these messages and show significant improvement. However, the presented results are conducted under simplistic propagation model and mobility model, which can lead to false conclusions. While other comparative studies use standard AODV with different set of parameters depending on the context of the paper, in [3] instead of creating a new routing algorithm to enhance AODV, the main contribution consists in improving the performance of the existing wellknown routing protocol by optimally tuning its parameters. The experiments reveal that the produced configurations of AODV significantly improve its performance. The mobility model and propagation model used are not depicted. Another main contribution to improve AODV (in a static environment) has been done by computing multiple paths in a single route discovery. Since each route discovery incurs high overhead and latency, the frequency of route discoveries must be kept low for ondemand protocols to be effective. Ondemand Multi path Distance Vector Routing in Ad Hoc Networks (AOMDV) depicted in [7] achieves this objective. Such as AODV, many works tried to enhance AOMDV. There exist many other protocols, and many comparisons research between protocols. Yet, to the best of our knowledge, no comparison used both realistic mobility model and propagation model. So, in order to analyze and test most of these proposed AODV enhancements, we use the well known Network Simulator 2 (NS2) [8] which has been improved by adding a realistic propagation tool together with a realistic mobility model. Simulation results are presented in this paper. The rest of this paper is organized as follows. Section II depicts AODV basics. Section III shows the impact of AODV parameters and explains how to optimally tune them according to related works [2] and [3]. Section IV depicts
AOMDV. Results obtained from NS2 realistic simulations are presented in Section V. Finally, Section VI concludes the paper. II.
AODV
AODV [1] is a singlepath, reactive routing protocol. Route discovery uses a Route REQuest (RREQ) – Route REPly (RREP) cycle. When a source node has data to send to a destination and does not know the route to this destination, it floods a RREQ message throughout the network. Several RREQ messages, each traveling on different paths, will reach intermediate nodes or the destination node. This node replies a RREP message only to the first received RREQ and drops subsequent RREQ with the same source sequence number and broadcast ID. The RREQ message that arrived at the earliest is likely to have traversed a path with low delay and/or hop count. Representing the weight of each link in the network by the delay incurred on the link, AODV reduces to finding a minimumweight path between the source and the destination. An important feature of AODV is the maintenance of timerbased states in each node. Each node broadcasts periodically a Hello message for local connectivity. The node assumes a link break to a neighbor node if it does not receive any message from a known neighbor node for some time. In this case, a Local Repair procedure is launched ; when the node that detects a link break is closer to the destination node than to the source node, it reinitializes a route discovery process to the destination node. The node buffers the arrival packets during the local repair. When the local repair is successful, the node starts sending the buffered data packets to the destination node via the new found route. If the local repair is unsuccessful, the node propagates a Route ERRor (RERR) message, which contains a list of unreachable destinations towards the source node. When the source node receives this RERR, the route to the destination is made invalid and it should be reestablished by a new discovery process. Another distinguishing feature of AODV is the ability to provide unicast, multicast and broadcast communication. AODV uses broadcast route discovery messages and then unicast route reply messages. In AODV, a routing table stores information about the next hop to the destination and uses a sequence number to guarantee loopfree routes. Based on the AODV protocol, one route entry is maintained in the routing table of each node for each destination that the node is communicating with in the network. One drawback of AODV that has been extensively focused is the single route abstraction. When a link failure is encountered on an active route, single route abstraction requires to start a new time consuming route discovery process. III.
AODV PARAMETERS OPTIMIZATION
A. Hello Messages The hello message not only continuously broadcasts to get clearer view of the local network vicinity, it also
produces some drawbacks for the network in general. Hello messages consume network resources and bandwidth just as other messages increasing interference and collision with data and other control messages. According to [2], optimization of the hello message timer may drastically enhance AODV performance. Depending on the speed and transmission power, the hello message interval should change in order to adapt to the network. The faster the network topology changes, the lower the hello interval should be. In this paper, we conducted simulations with different hello intervals in order to find the best hello interval time, which will adapt the network topology changes without increasing interference and collision. B. AODV Timers The performances of AODV is significantly influenced by the choice of its configuration parameters. Discovering the best values to assign to these parameters and understanding their impact on the network behavior trade off is still hard. Authors of [3] try to answer this challenge using several optimization strategies for AODV configuration in VANETs. Each method tries to optimize the Packet Delivery Ratio over other metrics. In this paper, we use 3 different sets of parameters in order to find the optimum set for our VANET scenarios. We run simulations with the set of the main AODV parameters as specified in RFC 3561, the set of parameters from the NS2.34 standard AODV implementation and a set of parameter based on the best optimization strategy proposed in [3]. This last set main goal is to improve the Packet Delivery Ratio over other metrics with a metaheuristic method. In the rest of this paper, this set of parameters will be called Tuning AODV. IV.
AOMDV
AOMDV protocol [7] is an extension based on AODV. AOMDV employs the “Multiple LoopFree and Link Disjoint path” technique. For route discovery, route request messages are propagated throughout the network thereby establishing multiple paths at destination node and at the intermediate nodes. Loop freedom is guaranteed using the notion of “advertised hop count” for a given destination sequence number (or RREQ message) at each node in the network. The “advertised hop count” of a node is basically the hop count incurred by the first RREQ packet for a given destination sequence number from the source to destination. When a node has no route to the destination, it forwards only the first arriving RREQ packet. When a node has a valid route to the destination and receives a duplicated RREQ, it checks whether the RREQ arrived on a new nodedisjoint path using flooding property. If so, the node checks whether the hop count incurred by this RREQ is less than that of the primary path. As the primary path is selected similar to the procedure used in AODV, it is most likely to be a minimum delay path, while the alternate paths have hop counts shorter or equal to the primary path. Thus, AOMDV could fit into the category of routing
protocols based on minimumweight path routing. Since AOMDV selects only linkdisjoint or nodedisjoint paths, the multiple paths are likely to have infrequent route discoveries at low mobility compared to singlepath AODV. In this paper, we conduct simulations with NS2 standard implementation of AOMDV in order to analyze the impact of multi paths into the VANET context. V.
SIMULATIONS SETUPS AND RESULTS
In order to analyze and compare the enhancements presented in this paper, we conduct simulations with the widely used NS2 simulator [8]. A. Simulations setups In VANET simulations, the use of unrealistic mobility models (e.g. constant speed, pause time method, etc.) and/or unrealistic propagation model (e.g. free space, tworay ground, etc.) will give results that are biased and rather optimistic. The influence of bad links is thus underestimated. Environment interactions affect the link quality significantly. Reference [9] highlights the overoptimistic results, when using freespace model propagation, compared to a deterministic raytracing model. In these results, while the free space model provides a 100% packet delivery ratio, with the use of a realistic propagation simulator called Communication Ray Tracer (CRT) this ratio falls down to 60%. When the average number of hops is 1.2 for the free space, the use of the raytracing simulator gives a minimum of 2.5 hops.
Parameter
Value
Network simulator
NS2
Simulation time
55 s
Number of nodes
60
Simulation area
Munich city center 1000m*1000m
Data Type
CBR
MAC Layer
802.11p
Communications
10 Simultaneous EmitterReceiver Transmissions
MAC Rate
6 Mbps
Propagation simulator
Communication Ray Tracer
Propagation interference
1 Reflexion / 1 Diffraction
Transmission power
100 mW
Mobility simulator
VanetMobiSim
Mobility model
Random trip & Intelligent driver model
Speed
Up to 16 m/s
Vehicles starting points and motion
Randomly seeded (30 seeds)
Table 1.Summary of simulation parameters All the results presented show the average results of 30 mobility scenarios. Each one has randomly distributed starting positions of nodes and motions around the Munich city center. B. Simulations results Figure 2 to Figure 4 show the impact of Hello interval variation in simulations with different mobilities, in terms of Packet Delivery Ratio (PDR), delay and Routing Overhead vs Hello Interval for the AODV protocol. All other AODV parameters are the one commonly used in NS2 (cf. Table 1). An adaptive algorithm has also been tested. The Hello Interval is adapted to speed according to the recommendation of [2]. When the speed increases, the Hello Interval decreases from 3 s to 0.5 s and vice versa.
Figure 1. Free Space propagation model (green) and CRT simulator (blue) used in the Munich city center environment.
Packet Delivery Ratio (%)
To experiment quality of service, simulation should compute correct attenuation and error rate by taking into account obstacles, multi paths effects, etc. All of our simulations were conducted using the ray tracing simulator presented in [9]. Mobility has been generated by VanetMobiSim [10]. Each simulation was conducted with 60 nodes, 10 simultaneous communications, a mobility up to 16 m/s and with a 55 seconds simulated time. All communications between nodes start and stop at the same time with 10 different emittersreceivers. The urban environment chosen for simulation is the Munich city center. Table [1] summarizes the chosen parameters.
PDR 50 40 30 20 10 0 0,5 0,1
1,5 1
3 2
Adaptive 30
Hello Interval (s) Figure 2. PDR vs Hello Interval
Figure 2 shows that the average PDR variation generated by the Hello message frequency is quite low. However, we notice 7% of received packet difference between common
Hello Interval parameter (1 sec) and extreme values like 0.1 s and 30 s. The impact of the Hello Interval on PDR is not very significant according to these simulation results. Figure 3 shows similarly to figure 2 that the Hello interval does not affect significantly AODV average results in term of delay except for the extreme 0.1 s value which has a bad delay compared to the standard Hello Interval value. Figure 4 confirms that the number of control packets increases as the Hello Interval decreases.
Delay (s)
Delay 8 7 6 5 4 3 2 1 0 0,5 0,1
1,5 1
3 2
Adaptive 30
Hello Interval (s)
Parameters used in our next set of simulations are RFC values, commonly used parameters in NS2, and tuned parameters according to [3] to fit VANET configurations. Table [2] summarizes all parameters sets. Parameter
RFC
NS2
Tuning
ACTIVE ROUTE TIMEOUT
3 s
10 s
1 s
ALLOWED HELLO LOSS
2
3
1
MY ROUTE TIMEOUT
6 s
10 s
2 s
NET DIAMETER
35
30
20
NODE TRAVERSAL TIME
0,04 s
0,03 s
0,01 s
NET TRAVERSAL TIME
2,8 s
1 s
0,4 s
RREQ RETRIES
2
3
1
RREQ RATELIMIT
10 s
10 s
1 s
TTL START
1 ms
5 ms
1 ms
TTL INCREMENT
2 ms
2 ms
1 ms
TTL THRESHOLD
7 ms
7 ms
1 ms
Table 2. A summarize of parameters used in simulations
Figure 5 to Figure 7 show average simulations results in term of Packet Delivery Ratio (PDR), average end to end delay and Routing Overhead for AOMDV and AODV according to different tuning parameters and different mobilities (cf. Section V.A).
Figure 3. Delay vs Hello Interval Routing Overhead
PDR
50000 40000 30000 20000 10000 0 0,5 0,1
1,5 1
3 2
Adaptive 30
Hello Interval (s) Figure 4. Routing Overhead vs Hello Interval
According to these simulation results and contrary to expected results from [2], the Hello Interval does not help to enhance the PDR and delay performance of AODV. The commonly used parameter of 1 s is well suited in our VANET context, and no other Hello Interval value improves significantly the network performance. However, a Hello Interval set to 2 s or 3 s may half the Routing Overhead with only a minor impact on PDR and delay. Results shows that an adaptive Hello Interval based on speed does not enhance the performance. Yet, a total suppression of the Hello messages may decrease performance as shown by the 30 s Hello Interval results. The Hello message also still remains useful in order to exchange information between nodes in the AODV protocol.
Packet Delivery Ratio (%)
Routing Overhead
60000 50 40 30 20 10 0 Standard
RFC
AOMDV
Tuning
AODV Parameters' set Figure 5. PDR of AODV and AOMDV
Figure 5 shows that Tuning AODV according to reference [3] gives the best results in terms of PDR which was expected as Tuning AODV tries to improve PDR over all other metrics. The RFC parameters seem not to be appropriate to the VANET context as they give the lowest PDR results. Both the Standard AODV parameters values and the AOMDV algorithm give interesting results, but still not as good as Tuning AODV. Both values do not diverge more than 6%, which means that our routing algorithms may improve PDR but not significantly. PDR results are still low in a VANET realistic context.
VI.
Delay
In this paper, we analyze under realistic conditions different AODV enhancement propositions to answer the routing challenges in the VANET context. Our results show that AOMDV gives the best results over other all methods. Yet, it may improve PDR, delay and Routing Overhead but the results are still poor (PDR lower than 50% and Average end to end delay over 1 sec) and not satisfying for good QoS communications. This domain is still quite active, and more investigations with newer algorithms or other methods should be done before being able to setup the best solution.
5
Delay (s)
4 3 2 1 0 Standard
RFC
AOMDV
Tuning
AODV Parameters' set Figure 6. Delay of AODV and AOMDV
Figure 6 shows that Tuning AODV also gives the best results in terms of average end to end delay. AOMDV also improves significantly the average delay compared to both RFC and standard AODV. It is quite surprising that tuned parameters also improve delay rather than only PDR because the tuning method focuses on PDR enhancement over other metrics. AOMDV shows interesting results in term of delay, and the impact of multi paths is clearly shown with these results. RFC parameters show one more time that they are not suitable for the VANET context. Routing Overhead 20000
Routing Overhead
CONCLUSION AND FUTURE WORKS
15000 10000 5000 0 Standard RFC
AOMDV Tuning
AODV Parameters' set Figure 7. Overhead of AODV and AOMDV
One can see in Figure 7 the drawback of the AODV tuned set of parameters. The PDR and average end to end delay may be improved but the Routing Overhead drastically increases. This may not be important in a small network with 60 moving nodes, but will become an important problem in networks involving more than 1000 cars. Tuned AODV fits best our VANET scenarios but may be quite unadapted to bigger networks. At the same time, AOMDV is also able to improve PDR and delay but decreases Routing Overhead too, which is very important. Other results do not show important differences in term of routing overhead.
ACKNOWLEDGMENT This work is supported by the French Région Poitou Charentes.
REFERENCES [1]
C. E. Perkins, E. M. BlkdingRoyer, S. Das “Ad Hoc Ondemand Distance Vector (AODV) Routing”, IETF Internet Draft, November 2001. [2] E. Natsheh, A. Jantan, S. Khatun, S. Subramaniam “Adaptive Optimizing of Hello Messages in Wireless AdHoc Networks” The International Arab Journal of Information Technology, Vol. 4, No. 3, July 2007. [3] J. GarciaNieto and E. Alba, “Automatic Parameter Tuning with Meta heuristics of the AODV Routing Protocol for Vehicular AdHoc Networks” in Applications of Evolutionary Computation, 2010. [4] M. Zapata and N. Asokan, “Securing Adhoc Routing Protocols,” in Proc. of ACM Workshop on Wireless Security (WiSe), Atlanta, GA,Sept. 2002. [5] J. Ledy, H. Boeglen, B. Hilt, A. Abouaissa, R. Vauzelle, "An enhanced AODV protocol for VANETs with realistic radio propagation model validation," in Intelligent Transport Systems Telecommunications, (ITST), Oct. 2009. [6] P. Fazio, F. De Rango, C. Sottile, P. Manzoni, C. Calafate, "A distance vector routing protocol for VANET environment with Dynamic Frequency assignment," IEEE Wireless Communications and Networking Conference (WCNC), March 2011 [7] M.K. Marina, S.R. Das, "Ondemand multipath distance vector routing in ad hoc networks," in Ninth International Conference on Network Protocols, Nov. 2001 [8] The Network Simulator Project Ns2, http://www.isi.edu/nsnam/ns/ [9] W. Hamidouche, R. Vauzelle, C. Olivier, Y. Pousset, and C. Perrine, “Impact of realistic mimo physical layer on video transmission over mobile ad hoc network,” in 20th Personal, Indoor and Mobile Radio Communications Symposium, 2009. [10] J. Haerri, F. Filali, and C. Bonnet, “Mobility models for vehicular ad hoc networks : a survey and taxonomy,” in IEEE communications surveys and tutorials, vol. 11, pp. 19–41, IEEEINST ELECTRICAL ELECTRONICS ENGINEERS INC, 2009.