On the performance of MANETs under different mobility patterns and routing protocols and its improvement based on fixed relay nodes Mouna Abdelmoumen1 , Imen Arfaoui1 , Mounir Frikha1 , Tijani Chahed2 1 Sup’com Route de Raoued Km 3,5 - 2083 El Ghazala Ariana, Tunisia {mouna.abdelmoumen, m.frikha}@supcom.rnu.tn,
[email protected] 2 Institut Telecom; Telecom SudParis; UMR CNRS SAMOVAR 9 rue C. Fourier - 91011 Evry CEDEX - France
[email protected] Abstract—Mobile Ad Hoc Networks (MANETs) are wireless networks where (user) nodes act themselves as relays to other nodes, resulting in a variable topology that follows the users mobility pattern. The latter can vary from one context to another resulting in different network performance, in terms of throughput for instance. This is the rationale behind the present work where we, first, quantify the performance of the network under different mobility patterns and routing protocols, and where we, second, propose the use of fixed relays so as to enhance the performance in case of ill-behaved mobility schemes. Index Terms—MANETs, mobility models, routing protocols, metrics.
I. I NTRODUCTION Mobile Ad Hoc Networks (MANETs) are wireless networks where (user) nodes act themselves as relays to other nodes. The performance in such a network, in terms of throughput for instance, is largely dependent on the varying topology of the network, which depends itself on the users mobility pattern. The latter can vary from one context to another, and can take different models, mainly due to the taking into account, or not, of spatial as well as temporal correlations between nodes movements. Some examples are Random Way Point (RWP) [1] (in this case, the node movement is free of restrictions, both temporal and spatial), Graph Based Mobility Model (GBMM) [2] (it is similar to RWP but it constrains the node movement to a connected graph), Smooth Random Mobility Model (SRMM) [3] (it adds a temporal dependency to the RWP node movement), and Manhattan Mobility Model (MMM) [4] (it includes all dependencies. It makes use of a map to confine movement to lanes. Moreover, nodes move according to a temporal correlation. The nodes speed is constrained by the speed of the front node in the same lane). Routing is yet another important factor that can impact the network performance. Several routing protocols have been proposed in the context of MANETs, the most widely referenced ones being: AODV [5], OLSR [6] and DSR [7]. The latter is a reactive routing protocol, where routes are created on demand using two mechanisms: route discovery to find routes and
route maintenance to preserve them. AODV works similarly to DSR using route discovery and maintenance mechanisms. It, however, uses hop by hop routing. OLSR is a proactive routing protocol, where the information about the network topology is exchanged by control packets (Hello messages). OLSR makes use of Multi-Point Relays (MPR) nodes to retransmit broadcast messages and hence reduce control packets. Several works studied mobility and its impact on MANET performance. Reference [8] for instance showed that the mobility of nodes increases the source-destination throughput. As of its relationship to routing, the study contained in [9] showed how mobility impacts the performance of reactive routing protocols in MANETs. Other works proposed enhancements to MANETs. The authors in [10] proposed new versions to the original OLSR routing protocol based on a specific mobility parameter called the degree of node mobility. In addition, the authors in [11] introduced mobility aware agents in ad-hoc network nodes using AODV protocol. Those agents select the best path between source and destination based on the node movement frequency. These works, however, did not give explicit details on the relationship between mobility and routing in studying the performance of MANETs. In addition, the proposed performance improvement was essentially done by adding modifications to the routing level, despite the fact that the reason of the poor performance was mobility. Our aim in this work is to, first, assess the performance of MANETs, in terms of joint mobility-routing considerations and metrics, and, second, in the case of poor performance, typically because of a degenerate mobility model which fractions the overall network into non-connected sub-networks, propose the use of fixed relays in order to maintain the connectivity of all different parts of the network. II. M OBILITY-ROUTING R ELATIONSHIP In order to study the network performance, in terms of throughput, we focus on the paths established between the source and the destination, and define metrics that quantify
978-1-4673-0229-6/12/$31.00 ©2012 IEEE
their effect on it. In order to do so, we first focus on the relationship between mobility and routing. Transmission between any two nodes i and j occurs only when these two nodes come within communication range, say r, of each other. Let us denote such a link by Lm (i, j). The path between a source and destination is a succession of such paths, whose creation/destruction is function of nodes encounters/dis-encounters, and hence mobility. Once this mobility-based path is established between the source and the destination, and before the effective transfer of information between source and destination, the routing protocol must exchange some information, such as routing table update, route request/response, etc, to let the source and destination nodes know about the existence of a path between them. A routing link between nodes i and j, denoted by Lr (i, j), will thus be created on top of the ”mobility” link, Lm (i, j). Both mobility and routing paths are composed of timevarying sub-, or unitary, paths relating successive nodes between the source s and destination d pairs. So, for the rest of the paper, in reason of the similarty between mobility and routing metrics and the paper space we fixe the index m or r for mobility or routing respectively to a . The mobility and routing unitary paths, denoted by upm and upr respectively, between nodes s and d are composed by k − 1 consecutive mobility or routing links, respectively, from s to d, and are defined by the same way as upa (s, d, tc , tv ) = {La (s, n1 ), ..., La (nk−2 , d)}. tc is the establishment or discovery time of the mobility or routing unitary path respectively. tv is the break or interruption time of the mobility or routing unitary path respectively. Also, we define ∀q, upaq (s, d) = upaq (s, d, tcq , tvq ). Then, the mobility and routing path, denoted by Pm and Pr respectively, between nodes s and d are composed by successive unitary paths as Pa (s, d, tc , tv ) = {upag (s, d), ..., upal (s, d)}. Pa (s, d, tc , tv ) is the mobility or routing path between s and d established or discovered respectively at time tc = tcg and broken interrupted respectively at time tv = tvl having ∀q ∈ [|g + 1, l|]; tcq tvq−1 or tcq − tvq−1 ≤ ε. The absence of mobility as well as routing paths APm and APr , respectively, between s and d corresponds to the absence of successive mobility and routing links respectively between s and d for a period of time larger than ε. So, we have, APa (s, d, ti , tf ) = (upai (s, d), upaf (s, d)) where tcf tvi .
The duration of the path absence is simply given by: AP Da (s, d, ti , tf ) = tf − ti
For an observation duration denoted by T = [tbegin , tend ], we define three sets. The first set, denoted by Pa (s, d, T ), contains all of the paths between s and d observed during interval T ; {Paz (s, d) = Pa (i, j, tcz , tvz ); tcz , tvz ∈ T, tcz ≤ tcz+1 ∀z 0}. The second set, denoted by APa (s, d, T ), contains all the absences of path between s and d observed during interval T ; {APaz (s, d) = APa (i, j, tiz , tfz ); tiz , tfz ∈ T, tiz ≤ tiz+1 ∀z 0}. The third set, denoted by M PSD , contains all source-destination pairs between which a mobility path will be investigated. For the observation duration T and M PSD set, we derive the following average metrics. The average path duration, P Da , is equal to : P Daz (s,d) Paz (s,d)∈Pa (s,d,T ) (s,d)∈M PSD
Card(Pa (s,d,T ))
Card(M PSD )
(s,d)∈M PSD
Card(APa (s,d,T ))
Card(M PSD )
(4)
where AP Daz (s, d) = AP Da (s, d, tiz , tfz ). Finally, Card is the number of elements in the set. IV. S IMULATION SETTING AND RESULTS In our study, we will consider the following mobility models: RWP, GBMM, SRMM and MMM. For RWP and SRMM mobility models, the value of the pause time is randomly chosen between 10 and 60s. For GBMM and MMM, we use the maps shown in Figures 1 (a) and (b), respectively. As of routing, we shall consider DSR, AODV and OLSR. Let the network be composed of 40 nodes, deployed over a simulation area equal to 800m×800m. The simulation duration is taken to be 1000s. Speeds of 1 and 20m/s are used to mimic the mobility of both pedestrians (low speed) and cars (high speed). Transmission ranges are equal to 100m. Let the traffic rate and the data packets size be 32kbits/s and 96Bytes, respectively. Traffic is generated during all simulation duration. Let M PSD be {(i, i + 20) ∀i ∈ [1, 20]}.
To quantify the mobility and routing paths, we define, first, the link duration, LDa (i, j, t), observed at time t, as the longest time interval, [t, t ], during which La (s, u) exists. Based on [9], we define the mobility and routing path duration, P Da (s, d, t1 , t2 ), which is equal to: min LDa (nh , nh+1 , tcq ) (1) 1≤h≤kq
where ∀ upaq (i, j); n1 = s, nkq = d and kq is the number of nodes of the unitary path.
(3)
where P Daz (s, d) = P Da (s, d, tcz , tvz ). The average of the path absence duration, AP Da , is equal to: AP Daz (s,d) APaz (s,d)∈APa (s,d,T )
III. M OBILITY AND ROUTING M ETRICS
upaq (s,d)∈Pa (s,d,tc ,tv )
(2)
Fig. 1.
Maps used for GBMM(a) and MMM (b)
Fig. 2.
Average path durations (a), Average of the path absence durations (b) and network throughput (c) for used mobility models and routing protocols.
Fig. 3. Simulation zones (a), type zone for RWP, GBMM and SRMM (b), type zone for MMM (c), relay nodes position for RWP, GBMM and SRMM at low speed (d) and high speed (e) and relay nodes position for MMM (f).
Fig. 4. Average path durations (a), Average of the path absence durations (b) and network throughput (c) for used mobility models and routing protocols without and with fixed relay nodes.
We generate 20 mobility scenarios for RWP, GBMM, SRMM and MMM based on [12], [13] and [14], respectively. Simulations are made using NS-2. Figure 2 shows, respectively, the average path durations, the average path absence durations and the throughput as a function of speed for the different mobility models and routing protocols stated above. As we can see, all metrics decrease with speed for all mobility models and routing protocols. This observation is due to the fact that the more nodes move quickly, the more node encounters happen, and this, for small durations. Hence, many path breaks and establishments (interruption and discovery of
routing path) occur. As a result, the path and absence of path durations decrease. Hence, we observe that the mobility path durations are better than the routing ones and, on the contrary, the absence of routing path durations are better than the mobility ones (Figure 2 (a) and (b)). The reason is that mobility paths persist more and are less sensitive to interruptions than the routing ones. In addition, Figures 2 (a) and (b) show that the path durations are too low in comparison with the absence of path durations. This is due to network fragmentation. In effect, as nodes move, in particular the source and destination nodes, they can go to isolated positions, and so, no communication
will be possible. In the present case, network fragmentation is frequent as path durations are too low compared to the absence of path durations. For the mobility models, MMM presents almost the smaller path durations and the larger absence of path durations. This fact is due to the MMM map shown in Figure 1(b). In effect, links are only formed if nodes move close to each other in the same or opposite lanes or at intersections. Unfortunately, those situations are less probable. RWP and SRMM have similar performances and are better than MMM. The reason is that RWP and SRMM have the same basic characteristics, they allow nodes to move in all direction, and so, links can be formed more frequently, allowing larger path durations. However, path durations are still too small in comparison to the absence of path durations. GBMM has the larger path durations and the smaller absence of path durations. Indeed, GBMM is, typically, RWP with graph restrictions. As nodes positions are fixed on the graph, as shown in Figure 1(a), the probability of link and path establishment is high. As of the routing protocols, OLSR has the larger values of the path and absence of path durations. DSR and AODV work similar. In effect, as OLSR is a proactive protocol, at each topology modification and/or periodically, messages are broadcasted to update network information. This fact increases the knowledge about the validity of paths. However, with a frequently changing network, these updates can increase the latency to detect the path interruption due to network congestion. On the contrary, DSR and AODV are reactive. And so, latencies characterize the routes discovery mechanism, but, due to the maintenance mechanism, path interruptions can be quickly detected. In particular, AODV presents larger values of the absence path durations than DSR. The reason is that DSR uses the MAC notification to detect link failure and AODV uses the periodic Hello messages which are broadcasted each 2s. And hence, DSR failure detection mechanism is more efficient than the AODV one. The throughput performance follows the mobility and routing metrics as explained before. Moreover, as can be seen from Figure 2(c), GBMM throughput is the largest for all routing protocols. AODV and DSR work better than OLSR. Those observations are due to the results of mobility and routing metrics obtained above. However, the obtained throughput is too low. In effect, the maximum that can be reached is equal to the source rate which is 32kbits/s. As shown, throughput is between about 2kbits/s (MMM/OLSR/20m/s case) and about 9kbits/s (GBMM/AODV/1m/s case). This low performance is caused by the network fragmentation as explained above. And so, this fact must be handled so as to improve the throughput. And this is the topic of the next section. V. P ROPOSAL As shown in the previous section, when the network fragmentation is frequent, paths cannot be available for a large duration and the network performs poorly. When mobility offers more path establishment opportunities, as in the case of GBMM, throughput achieves however a better performance.
We, then, propose, in this section, a network solution to improve the mobility metrics by offering new opportunities to establish paths and to preserve them. The solution is based on the use of additional fixed relay nodes. By this means, the duration of formed paths can be improved. The number and positions of these fixed nodes will be determined and optimized in such a way so as to reduce network fragmentation by relating the isolated nodes to the rest of the network, as will be seen next. A. Description of Proposal The number of the fixed relay nodes must be sufficient enough so as to improve the network performance, but must not exceed a certain limit so as not to overload the network. In addition, in order to fix the positions of the fixed relay nodes, we need to have an idea about the network instantaneous topology. In effect, this level of knowledge is essential to learn about possible network fragmentation and to fix the nodes where it is more suitable. To do this, we propose to divide the network area into smaller areas called simulation zones. For each simulation zone, we compute some metrics based on the nodes visits to the zone. We define the visit, denoted by V (i, j, ta , td ), which happens when node j goes to simulation area i at time ta and goes away from it at time td . And so, we have, for the observation duration T , the visit set to simulation zone i, denoted by Vi , as Vi = {V (i, j, taz , tdz ); taz , tdz ∈ T, taz ≤ taz+1 }. For each simulation zone i, we define three statistical metrics. The number of visits, N Vi equals to card(Vi ). The N Vi
average visit duration, DVi , equals to
tdz −taz N Vi
z=1
N Vi −1
. The averta
−ta
z z+1 age nodes inter-arrival time, iai , equals to z=1 N Vi −1 . Having the pervious metrics, we can define four types of zones. The desert zone are visited by few nodes, at dispersed times, which spend in it a little period of time (small values of N Vi and DVi and large ones for iai ). The home zone are visited by few nodes, at dispersed time, which spend in it a larger period of time (small values of N Vi and large ones for DVi and iai ). The crossing zone is visited by a lot of nodes, at close times, which spend in it a little period of time (large values of N Vi and small ones for DVi and iai ). The attractive zone is visited by a lot of nodes, at close time, which spend in it a larger period of time (large values of N Vi and DVi and small ones for iai ). With the definition of the simulation zone types, we can predict the relay nodes positions. As network fragmentation results in isolation of nodes from the rest of the network, these isolated nodes are typically in desert or home zones and the rest of nodes will be in crossing or attractive zones. And so, in order to relate the isolated nodes to the rest of nodes, fixed relay nodes must be positioned so as to allow connection between the two sets of zones: desert, home and crossing, attractive.
B. Performance of Proposal To implement our proposal in the studied network, we divide the simulation area into sixteen zones as shown in
Figure 3(a). We choose the number of fixed relay nodes to be eight as this number is not too large in comparison with the number of mobile nodes and not too small to serve our puspose. We compute the statistical zone metrics and deduce the type of each zone as shown in Figures 3(b) and (c). As RWP, SRMM and GBMM have similar results for the zone type, we represent them by the same figures. For RWP, SRMM and GBMM, isolated nodes positions are in the zones around the central ones (zones: 5,6,9,10). The rest of the nodes pass through the central zones. And so, at low speed, we must cover more the central zones than the surrounding ones. On the contrary, at high speed, we must cover more the home zones than the crossing ones. In effect, the more nodes present in central zones, the longer the path durations. When zones are crossing ones, path quality decreases. And so, it is more appropriate that fixed relay nodes cover the most stable zones in each case. Moreover, in order to be fair, we propose fixed relay nodes positions for RWP, SRMM and GBMM as shown in Figures 3(d) and (d). For MMM, as we are constrained by the used map (shown in Figure 1(c)), we choose to fix relay nodes around the central zones so as to allow connections between spaced lanes, as shown in Figure 3(f). Figure 4 shows the mobility and routing metrics and the network throughput before and after the use of the fixed relay nodes. For comparison, we also show the old values of mobility metrics and network throughput. As shown in Figure 4(a), the mobility path duration increases and the absence of mobility path duration decreases with the use of the fixed relay nodes, especially at low speed. This result is trivial because the additionnal fixed nodes offer more stability, in terms of connectivity, to the network. In effect, the relay nodes increase the mobility path establishment probability and its duration too, and so, they decrease the absence of mobility path duration. Moreover, at the same speed, we observe that the routing path durations are less than the mobility ones and not as was the case without the use of the fixed relay nodes (see Figure 2 (a)). This observation is the result of the network congestion. In effect, as the fixed relay nodes improve the connectivity of the network, paths between sources and destinations are more available and for a larger period of time. And so, having many connections to handle, the routing level may become congested and cannot take advantage from the availability of the mobility paths. On the other hand, Figure 4(b) shows that the absence of routing path duration are less than the mobility ones and not as without the use of the fixed relay nodes (see Figure 2 (b)). The reason is that the routing level becomes more stable and can overcome the network disconnection problems caused by the increase of path estbalishment probability. At high speed, the mobility and routing metrics do not yield a large improvement because nodes move too quickly causing a high connection/disconnection frequency despite of the use of the fixed relay nodes. As of throughput, Figure 4(c) shows that throughput in-
creases with the use of the fixed relay nodes for all mobility models and routing protocols. The reaseon is that packets are lost in smaller number because the absence routing path duration is less than without the use of the fixed relay nodes. In addition, as explained in section IV, DSR and AODV work better than OLSR. For MMM, we have the best throughput improvement, around 50%. This is due to the good positioning of the fixed relay nodes which cover large parts of the nodes trajectory (see Figure 3(e)). VI. C ONCLUSION We have shown in this work how mobility, combined with routing, impacts the MANET network performance, in terms of throughput for instance, due mainly to links and paths establishment and breaks. Ultimately, mobility may result in fragmentation of the network and hence very poor performance. To remedy to this, we have proposed, and optimized, the use of fixed relay nodes which would maintain the overall network connectivity and improve hence the overall throughput performance. R EFERENCES [1] D.B. Johnson and D. A. Maltz, ”Dynamic source routing in ad hoc wireless networks”, In Mobile Computing, Kluwer Academic Publishers, 1996. [2] J. Tian, J. Hahner, C. Becker, I. Stepanov and K. Rothermel, ”Graphbased mobility model for mobile ad hoc network simulation”, SS’02, San Diego, Apr. 2002. [3] C. Bettstetter, ”Mobility modeling in wireless networks: categorization, smooth movement and border effects”, ACM Mobile Computing and Commu. Rev., Jul. 2001. [4] F. Bai, N. Sadagopan, and A. Helmy, ”A framework to systematically analyze the impact of mobility on performance of routing protocols for adhoc networks”, INFOCOM ’03, San Francisco, Mar.-Apr. 2003. [5] C. Perkins, E. Belding-Royer and S. Das, ”Ad hoc On-Demand Distance Vector (AODV) Routing”, RFC 3561 Experimental, Jul. 2003. [6] T. Clauser and P. Jacquet, ”Optimized link State Routing Protocol (OLSR)”, RFC 3626 Experimental, Oct. 2003. [7] D. Johnson, Y. Hu and D. Maltz, ”The Dynamic Source Routing Protocol (DSR) for Mobile Ad Hoc Networks for IPv4”, RFC 4728 Experimental, Feb. 2007. [8] M. Grossglauser and D. Tse, ”Mobility increases the capacity of ad hoc wireless networks”, IEEE/ACM Trans. Netw., vol. 10, no. 4, pp. 477486, Aug. 2002. [9] F. Bai, N. Sadagopan, B. Krishnamachari and A. Helmy, ”Modeling path duration distributions in manets and their impact on reactive routing protocols”, IEEE Journal on Selected Areas in Communications, 22(7):13571373, Sept. 2004. [10] N. Enneya, K. Oudidi, J. Oubaha, and M. El Koutbi, ”Enhancing Delay in MANET Using OLSR Protocol”, International Journal of communications, Network and System Sciences IJCNS”, vol. 2, No. 4, 2009. [11] Idrees, M. Yousaf, M.M. Jaffry, S.W. Pasha, M.A. Hussain, S.A ”Enhancement in AODV Routing Using Mobility Aware Agents” . IEEE - 2005 International Conference on Emerging Technologies. September 17-18, Islamabad. [12] R. Baumann, F. Legendre and P.Sommer, ”Generic mobility simulation framework (GMSF)” , Mobility Model,08 : Proceeding of the 1st ACM SIGMOBILE workshop on Mobility models, Hongkong, China. [13] The CANU Mobility Simulation Environment (CanuMobiSim) [Online]. Available: http://canu.informatik.uni-stuttgart.de/mobisim/ [14] The mobility tool generators [Online]. Available: http://nile.cise.ufl.edu/important/software.htm
