Displacement-based Route Update Strategies for Proactive Routing Protocols in Mobile Ad Hoc Networks

Transcription

1 Displaement-based Route Update Strategies for Proative Routing Protools in Mobile Ad Ho Networks Mehran Abolhasan 1 and Tadeusz Wysoki 1 1 University of Wollongong, NSW 2522, Australia mehran@titr.uow.edu.au, wysoki@uow.edu.au Abstrat This paper presents a new route update strategy for performing proative route disovery in Mobile Ad ho Networks (MANETs). In this strategy, the rate at whih route updates are sent into the network is ontrolled by how often a node hanges its loation by a required distane. We refer to this updating strategy as Minimum Displaement Update Routing (MDUR). We implemented MDUR on top of the Fisheye State Routing () protool and investigated its performane by simulation. The simulations where performed in a number of different senarios, with varied network mobility, density, traffi and boundary. Our results indiate that MDUR has lower levels of ontrol overhead than and ahieves higher levels of throughput as the density and the level of traffi in the network is inreased. I. INTRODUCTION Mobile Ad Ho Networks (MANETs), are made up of a number of nodes, whih are apable of performing routing without using a dediated entralised ontroller or a base station. This key feature of these networks enable them to be employed in plaes where an infrastruture is not available, suh as in disaster relief and on battle grounds. However, the dynami nature of these networks and the sarity of bandwidth in the wireless medium, along with the limited power in mobile devies (suh as PDA s or laptops) makes routing in these networks a hallenging task. A Routing protools designed for MANETs must work onsistently as the size and the density of the network varies and effiiently use the network resoures to provide eah user with the required levels of quality of servie for different types of appliations used. With so many variables to onsider in order to design an effiient routing protool for MANETs, a number of different types of routing strategies have been proposed by various authors. These protools an be lassified into three groups; global/proative, on-demand/reative and hybrid. Most proative routing protools are based on the link state and distane vetor algorithms. In these protools, eah node maintains upto-date routing information to every other node in the network by periodially exhanging distane vetor or link state information using different updating strategies (disussed in the following setion). In on-demand routing protools eah node only maintains ative routes. That is, when a node requires a route to a partiular destination, a route disovery is initiated. The route determined in the route disovery phase is maintained while the route is still ative (i.e. the soure has data to send to the destination). The advantage of on-demand protools is that they redue the amount of bandwidth usage and redundany by determining and maintaining routes when they are required. These protools an be further lassified into two ategories: soure routing and hop-by-hop routing. In Soure routed on-demand protools [12][18], eah data pakets arry the omplete soure to destination address. Therefore, eah intermediate node forwards these pakets aording to the information kept in the header of eah paket. This means that the intermediate nodes do not need to maintain up-to-date routing information for eah ative route in order to forward the paket towards the destination. Furthermore, nodes do not need to maintain neighbour onnetivity through periodi beaoning messages. The major drawbak with soure routing protools is that in large networks they do not perform well. This is due to two main reasons; firstly as the number of intermediate nodes in eah route grows, then so does the probability of route failure. To show this let P (f) = a.n, where P (f) is the probability of route failure, a is the probability of a link failure and n is the number of intermediate nodes in a route. From this 1, it an be seen that as n, then P (f). Seondly, as the number of intermediate nodes in eah route grows, then the amount of overhead arried in eah header of eah data paket will grow as well. Therefore, in large networks with signifiant levels of multihoping and high levels of mobility, these protools may not sale well. In hop-by-hop routing (also known as point-to-point routing) [6], eah data paket only arries the destination address and the next hop address. Therefore, eah intermediate node in the path to the destination uses its routing table to forward eah data paket towards the destination. The advantage of this strategy is that routes are adaptable to the dynamially hanging environment of MANETs, sine eah node an update its routing table when they reeiver fresher topology information and hene forward the data pakets over fresher and better routes. Using fresher routes also means that fewer route realulations are required during data transmission. The disadvantage of this strategy is that eah intermediate node must must store and maintain routing information for eah ative route and eah node may require to be aware of their surrounding neighbours through the use of beaoning messages. Hybrid routing protools have been proposed to inrease the salability of routing in MANETs [11][19][13][9] [15]. These protools often an behave reatively and proatively at different times and they introdue a hierarhial routing struture to the network to redue the number of retransmitting nodes during route disovery or topology disovery. Eah node periodially maintains the nearby topology by employing a proa- 1 Assuming that the intermediate nodes have a probability of a link failure of a > 0

2 tive routing strategy (suh as distane vetor or link state) and maintain approximate routes or on-demand routes for far away nodes. In this paper, we propose a new route updating strategy to perform proative route disovery in mobile ad ho networks. In this strategy, the rate at whih route updates are sent is ontrolled by the rate of displaement of eah node. This is determined by using the servies of a GPS. In [3], we briefly mentioned MDUR, in this paper we give a full desription of this strategy, and investigate its performane under different network senarios using a simulation tool.the rest of this paper is organised as follows. In setion II, we desribe a number of different route update strategies proposed in the literature. Setion III desribes our route updating strategy. Setion I desribes the simulation environment, parameters and performane metri used to investigate the performane of our route update strategy. Setion presents the disussion of our simulation results and setion I presents the onlusions of the paper. II. RELATED WORK Proative route disovery provides pre-determined routes for every other node (or a set of nodes) in the network at every node. The advantage of this is that end-to-end delay is redued during data transmission, when ompared to determining routes reatively. Simulation studies [5][8][4],whih have been arried out for different proative protools show high levels of data throughput and signifiantly less delays than on-demand protools (suh as DSR) for networks made up of up to 50 nodes with high levels of traffi. Therefore, in small networks using real-time appliations (e.g. video onferening), where low end-to-end delay is highly desirable, proative routing protools may be more benefiial. In this setion, we desribe a number of different route update strategies proposed in the literature to perform proative routing. A. Global updates Proative routing protools using global route updates are based on the link state and distane vetor algorithms, whih were originally designed for wired networks. In these protools, eah node periodially exhanges its routing table with every other node in the network. To do this, eah node transmits an update message every T seonds. Using these update messages, eah node then maintains its own routing table, whih stores the freshest or best route to every known destination. The disadvantage of global updates is that they use signifiant amount of bandwidth. Sine they do not take any measures to redue ontrol overheads. As a result data throughput may suffer suffer signifiantly, espeially as the number of nodes in the is inreased. Two suh protool are DSD [16] and WRP [14]. B. Loalised updates To redue the overheads in global updates, a number loalised updating strategies where introdued in protools suh as GSR[10] and [8]. In these strategies, route update propagation is limited to a loalised region. For example, in GSR Fig. 1. J I D X C Illustration of the fisheye sope in S A B E T G H F Hop = 1 Hop = 2 Hop = 3 eah node exhanges routing information with their neighbours only, thereby eliminating paket flooding methods used in the global routing. is a diret desendent of GSR. This protool attempts to inrease the salability of GSR by updating the nearby nodes at a higher frequeny than the node whih are loated far away. To define the nearby region, introdues the fisheye sope (as shown in Figure 1). The fisheye sope overs a set of nodes whih an be reahed within a ertain number of hops from the entral node shown figure 1. The update messages with greater hop ounts are sent at a lower frequeny. This redues the auray of the routes in remote loations, however, it signifiantly redues the amount of routing overheads disseminated in the network. The idea behind this protool is that as the data pakets get loser to the destination the auray of the routes inreases. Therefore, if the pakets know approximately what diretion to travel, as they get lose to the destination, they will travel over a more aurate route and have a high hane of reahing the destination. C. Mobility based updates Another strategy whih an be used to redue the number of update pakets is introdued in DREAM [4]. The author proposes that routing overhead an be redued by making the rate at whih route updates are sent to the speed at whih eah node travels. Therefore, the nodes whih travel at a higher speed disseminate more update pakets than the ones that are less mobile. The advantage of this strategy is that in networks with low mobility this updating strategy may produe fewer update pakets than using a stati update interval approah suh as DSD. Similar to, in this protool, updates are sent more frequently to nearby nodes than the ones loated far away. D. Conditional or Event-driven updates The number of redundant update pakets an also be redued by employing a onditional (also known as event-driven) based update strategy [16][7]. In this strategy a node sends an update if ertain different events our at any time. Some events whih an trigger an update is when a link beomes invalid or when

3 a new node joins the network (or when a new neighbour is deteted). The advantage of this strategy is that if the network topology or onditions are not hanged, then no update pakets are sent. Therefore, eliminating redundant periodi update dissemination into the network. III. PROPOSED STRATEGY In this setion, we propose Minimum Displaement Update Routing (MDUR). This route update strategy attempts disseminate route update paket into to the network when they are required rather than using purely periodi updates. This is ahieved by making the rate at whih updates are sent proportional to the rate of displaement. That is, the more a node hanges loation by a threshold distane the more updates are transmitted into the network. The rate of displaement an be measured using a Global Positioning System (GPS). A. Overview and Definition The idea behind this strategy is to redue the amount of periodi route updates by restriting the update transmission to nodes whih satisfy the following onditions: 1) A node experienes or reates a signifiant topology hange. 2) A node has not updated for a minimum threshold time. In the first ondition we assume that a node experienes a signifiant topology hange if it has migrated by a minimum distane from one loation to another loation. By migrating from one loation to another the routes onneted to the migrating node (and the route to the migrating node itself) may signifiantly hange. Therefore, the migrating node is required to transmit an update paket through the network (or parts of the network) to allow for realulation of more aurate routes. To illustrate how MDUR works, suppose node S (see figure 2) migrates from one loation to another as shown. From this migration it an be seen that the neighbour topology of node S has hanged, whih has also signifiantly hanged the topology of the network. Therefore, the dissemination of an update paket at this time will be benefiial as eah node in the network an rebuild their routing tables and store more aurate routes. B. Desription of MDUR Algorithm With MDUR, eah node starts by reording its urrent loation and sets it as its previous loation. They will also reord their urrent veloity and time. Using this information eah node determines when the next update should be sent. When this update time is elapsed the nodes hek to see if their migration distane is greater than the required threshold distane. If yes, an update is sent. Otherwise, no update is sent and the next update time is estimated aording to the urrent loation and veloity of the node. If the urrent veloity is zero the node an assume a maximum veloity or set a minimum wait time aording to an update time onstant, whih has been used in the MDUR algorithm. The MDUR algorithm is outlines below. Algorithm MDUR ( The MDUR algorithm ) 1. L p Previous loation 2. L Current loation Fig. 2. A C B D S i F (a) Illustrating Initial position for node S A C B D F H S f (b) Illustrating final position for node S Illustration of node migration in MDUR 3. L p L 4. D T The threshold distane 5. Disseminate update paket 6. speed of node 7. T urrent time 8. τ ( D T 9. while (node is online) 10. wait until T = τ 11. L urrent loation 12. if (dist(l, L p) D T 13. Disseminate update paket 14. L p L 15. τ ( D T 16. else 17. τ ( D T dist(l,l p) Displaement updates are more benefiial than using updates based purely on mobility (i.e. speed [4]). This is beause this strategy attempts to send an update when a topology hange ours. To show this, suppose node S (figure 2) moves rapidly towards node A for a short time suh that dist(l, L p ) < D T. Furthermore, it moves in suh a way that it maintains its links to nodes B and D. Now, assuming that there are no interferene during this time and nodes A, B and D stay stationary, then the topology of node S will not hange. Therefore, an update is not required in this network. However, in this ase a purely mobility based strategy suh as in [4] an update may be disseminated and it may ontinue to sent updates even if node S moves bak and forward between these two point. On ontrary, in this senario in MDUR no updates will be sent. C. Implementation Deision To evaluate the performane of our route updating strategy, we implemented MDUR on top of, whih we refer to as Hierarhial MDUR (). Reall that disseminates two types of update paket; Intrasope update pakets whih propagate within the fisheye sope and Intersope pakets whih propagate through the entire network. Therefore, H E E G G J J

4 TABLE I FISHEYE STATE ROUTING SIMULATION PARAMETERS Number of sopes 2 Intrasope update interval 5S Intersope update interval 15S Neighbour timout interval 15S TABLE II HIERARCHICAL MDUR SIMULATION PARAMETERS Number of sopes 2 Intrasope max timeout interval 10S Intersope max timeout interval 30S Minimum intrasope migration 30M Minimum intersope migration 200M we introdued two types of displaement updates; one for the intrasope and one for the intersope, and we modified the MDUR algorithm to disseminate these two updates. To initiate eah of these updates we also used two different threshold distanes; D intra and D inter for the intrasope and intersope updates respetively. To initiate the intrasope updates more frequently than intersope updates, we set D intra to be signifiantly less than D inter. Table I and II illustrate the parameters used in 2 and. The algorithm is outlined below. Algorithm ( The algorithm ) 1. L intra Loation at last inter-update 2. L inter Loation at last inter-update 3. L Current loation 4. L intra L 5. L inter L 6. D intra The intrasope threshold distane 7. D inter The intersope threshold distane 8. Disseminate intrasope update paket 9. Disseminate intersope update paket 10. speed of node 11. T urrent time 12. τ intra ( D intra 13. τ inter ( D inter 14. while (node is online) 15. wait until a timer expires 16. if (τ intra = expired) 17. if (dist(l, L intra) D intra 18. Disseminate intrasope update 19. L intra L else τ intra ( D intra 22. τ intra ( D intra dist(l,l intra ) 23. if (τ inter = expired) 24. if (dist(l, L inter) D inter 25. Disseminate intersope update 26. L inter L else τ inter ( D inter 29. τ inter ( D inter dist(l,l inter ) 2 The parameters were set aording to the ietf internet draft number 3 I. SIMULATION MODEL The aim of our simulation studies were to investigate the performane of our route update strategy under different levels of node density, traffi, mobility and network boundary. We simulated -DU and for eah senario in order to differentiate their performane. The simulations parameters and performane metris are desribed in the following setions. A. Simulation Environment and Senarios The Glomosim simulation tool was used to arry out our simulations [1]. GloMoSim is an event driven simulation tool designed to arry out large simulations for mobile ad ho networks. Our simulations were arried out for 50 and 100 node networks, migrating in a 1000m x 1000m boundary. IEEE DSSS (Diret Sequene Spread Spetrum) was used with maximum transmission power of 15dbm at 2Mb/s data rate. In the ma layer IEEE was used in DCF mode. The radio apture effets were also taken into aount. Two-ray path loss harateristis was used for the propagation model. The antenna hight is set to 1.5m, the radio reeiver threshold is set to -81 dbm and the reeiver sensitivity was set to -91 dbm aording to the Luent s wavelan ard[2]. A random way-point mobility model was used with the node mobility ranging from 0 to 20m/S and pause time varied from 0 to 900S. The simulation was run for 900S for 10 different values of pause time and eah simulation was averaged over five different simulation runs using different seed values. Constant Bit Rate (CBR) traffi was used to establish ommuniation between nodes. Eah CBR paket was 512 Bytes, the simulation was run for 10 different lient/server pairs and eah session was set to last for the duration of the simulation. B. Performane Metris To investigate the performane of the routing protools the following performane metris were used: Paket Delivery Ratio (PDR): Ratio of the number of paket sent by the soure node to the number of pakets reeived by the destination node. Normalised routing overhead (O/H): The amount of routing overhead transmitted through the network for eah data paket suessfully delivered to the destination. End-to-End Delay: The average end to end delay for transmitting one data paket from the soure to the destination The first metri is used to investigate the levels of data delivery (data throughput) ahievable eah protool under different network senarios. The seond metri will illustrate the levels of routing overhead introdued. The last metri ompares the amount of delay experiened by eah data paket to reah their destination.. SIMULATION RESULTS This setion presents our simulation results. The aim of this simulation analysis is to ompare the performane of HMUR with under different network senarios.

5 Pakets rev / Pakets sent (%) Paket Devlivery ratio (vs) Pause Time 0 Normalized O/H (bytes) Normalized O/H (vs) Pause Time 200 Fig. 3. PDR for 10S and 50N Fig. 5. Normalised O/H for 10S and 50N Pakets rev / Pakets sent (%) Paket Devlivery ratio (vs) Pause Time 0 Normalized O/H (bytes) Normalized O/H (vs) Pause Time 2000 Fig. 4. PDR for 10S and 100N Fig. 6. Normalised O/H for 10S and 100N A. Paket Delivery Ratio The graphs in Figure 3 and 4 illustrate the PDR results obtained for the 1000m x 1000m boundary. In the 50 node senario, and show similar levels of PDR However, in the 100 node network senario, starts to outperform. This is beause still maintains a similar level of PDR as in the 50 node senario, whereas has shown a signifiant drop in performane when ompared to the 50 node senario. This drop in performane is evident aross all different levels of pause time. This is beause under high node density the periodi updating strategy in starts to take away more of the available bandwidth for data transmission than HM- DUR. Furthermore, more updates may inrease hannel ontention, whih an result in more pakets being dropped at eah intermediate node. B. Normalised Control Overhead The graphs in Figures 5 and 6 illustrate the Normalized routing overhead experined in the 1000m x 1000m boundary. In our simulation, the maximum update intervals for the intrasope and intersope is set to be half of that of. Therefore, under high mobility (i.e. 0 pause time) if purely periodi updates where used in, the routes produed would have been less aurate, whih may have resulted in a drop in throughput. However, adapting the rate of updates by eah node to the rate of its displaement, allows the nodes to send more updates when they are required (i.e. during high mobility). This means that the auray of the routes will be high during high mobility where nodes are more likely to migrate more frequently and when mobility is low less updates are sent. From the results shown in Figures 5 and 6 it an be seen that produes less overhead than, aross all different levels of pause time and node density. C. Delays The graphs in Figure 7 and 8 illustrate the end-to-end delay experined in the 1000m x 1000m boundary. These results show that in eah data paket experines lower endto-end delay than in. The lower delay experiened is due to the higher level of aessability to the wireless medium. This is beause in eah node generates less route updates than it, whih means there is less ontention for the hannel when a data paket is reeived. Therefore, eah node an forward the data paket more frequently. I. ALTERNATE STRATEGIES AND IMPROEMENTS One way to inrease the salability of proative routing protools is by maintaining approximate routes to eah destination rather than exat routes. In [8] and [4], eah node maintains approximate (or less aurate) information to far away destinations, sine the updates from far away nodes are reeived less frequently. Similarly, in, nodes maintain approximate routing information to nodes loated far away by using the intersope displaement metri. In this setion, we propose

6 End to End Delay (s) Fig. 7. End to End Delay (s) End to End Delay (vs) Pause Time Fig. 8. Delays O/H for 10S and 50N End to End Delay (vs) Pause Time 0 Delays O/H for 10S and 100N a number of alternative strategies and improvements for MDUR and HMUR based on approximate routing. A. Minimum Topology Change Updates Another way to determine if an update is required is by monitoring the nearby topology and disseminating update pakets only when a minimum level of topology hange ours.to do this we introdue Minimum Topology Change Updates (MTCU). This strategy assumes that eah node maintains an intrasope and intersope topology like. However, instead of using purely periodi updates, the rate at whih updates are sent is proportional to a topology metri. MTCU is made up of two phases: these are startup phase and maintenane phase. The start up phase is initiated when a node enters the network (or when it omes online). During this phase, eah node starts by reording its loation and sends three updates, whih are: neighbour update, intrasope update and intersope update. Eah node then ounts the number of neighbouring nodes and the number of nodes in their intrasope. During the maintenane phase, the neighbouring topology is periodially monitored and the number of hanges is reorded. These hanges an inlude: disovery of new neighbour or the loss of a link. If a signifiant hange in the neighbouring topology is experiened an intrasope update is sent. Furthermore, eah node monitors its intrasope topology and ount the number of hanges, suh as the number of nodes in the intrazone and the number of route hanges for eah destination. If the intrasope has hanged signifiantly then an intersope update is sent. Note that the eah node maintains its neighbour onnetivity through beaoning messages. However, the rate at whih intrasope and intersope updates are disseminated is dependent on the rate at whih neighbouring or intrasope topology hanges, and periodi updates an be used only if eah node has not sent an intrasope or intersope update for long time 3. Therefore, reduing the number of redundant updates if no hanges our. This also means that fewer periodi updates maybe transmitted when ompared to protools whih use a purely periodi update strategy (suh as ). To detet if a signifiant neighbour or intrasope topology hange has ourred a topology metri an be used. In this ase, two topology metris are required to be kept, one for the neighbouring topology and one for the intrasope topology. The topology metri ounts the number hanges after the startup phase and triggers an update event if a ertain number of hanges our. The MTCU algorithm is outlines below. Note that the algorithm only shows the maintenane phase of MTCU. Algorithm MTCU ( The MTCU algorithm ) 1. NT Total urrent number of neighbours 2. NT p Total previous number of neighbours 3. T Total number of destinations in the intrasope 4. T p Total intrasope destinations previously reorded 5. N Total intrasope destinations previously reorded 6. P N hange perentage of neighbour hange required 7. P T hange perentage of topology hange required 8. N hange neighbour hanges reorded 9. T hange Topology hanges reorded 10. while (node is online) 11. wait for an update 12. if (update = neighbour) 13. update neighbour table 14. NT total number of neighbours 15. N hange + = number of hanges 16. if (N hange P N hange NT p) 17. Disseminate intrasope update 18. NT p NT 19. N hange if (update = Intrasope) 21. update topology table 22. T total number of neighbours 23. T hange + = number of hanges 24. if (N hange P T hange T p) 25. Disseminate intersope update 26. T p T 27. T hange if (update = Intersope) 29. update topology table In the above algorithm, the rate at whih updates are sent also depends on the perentage of hanges experiened (i.e. P T hange and P N hange ). The perentage of hange value an be a stati parameter between 0 and 100% and preprogrammed into eah devie. However, it maybe benefiial to dynamially hange its value aording to the network onditions. One way to do this is by estimating the available bandwidth at eah node and also for the intrasope, then varying the perentage hange values aording to the level of available bandwidth. Therefore, in times where the level of traffi (e.g. data and ontrol) is low, more updates an be sent to inrease the auray of the routes. 3 That is, when the network is stati then updates are sent at a lower frequeny when ompared to purely periodi updates

7 II. CONCLUSIONS This paper presents a new proative route update strategy for mobile ad ho networks. We present Minimum Displaement Update Routing (MDUR) and Hierarhial MDUR (). In these strategies, the rate at whih route updates are sent is proportional to the rate at whih eah node hanges its loation by a threshold distane. We implemented and ompared its performane with. Our results indiate that produes fewer routing overheads than while maintaining high levels of data throughput aross different network senarios. Furthermore, the results show that when the node density is high, reduing routing overhead an result in higher levels of data paket delivery and lower end-to-end delay for eah paket. In the future, we plan to simulate MDUR and with a simple geographi data forwarding (suh as those those desribed in [17]) and ompare its performane with shortest path routing. [19] Seung-Chul Woo and Suresh Singh. Salable Routing Protool for Ad Ho Networks. aepted for publiation in Journal of Wireless Networks (WINET). REFERENCES [1] Glomosim salable simulation environment for wireless and wired network systems. In [2] Orinoo p ard. In [3] M. Abolhasan, T. Wysoki, and E. Dutkiewiz. Salable Routing Strategy for Dynami Zone-based MANETs. In Proeedings of IEEE GLOBE- COM, Taipei, Taiwan, November [4] S. Basagni, I. Chlamta,.R. Syrotivk, and B.A. Woodward. A Distane Effet Algorithm for Mobility (DREAM). In Proeedings of the Fourth Annual ACM/IEEE International Conferene on Mobile Computing and Networking (Mobiom 98), Dallas, TX, [5] J. Broh, D. A. Maltz, D. B. Johnson, Y-Ch Hu, and J. Jetheva. A Performane Comparison of Multi-Hop Wireless Ad Ho Network Routing Protools. In Mobile Computing and Networking, pages 85 97, [6] S. Das, C. Perkins, and E. Royer. Ad Ho On Demand Distane etor (AOD) Routing. In Internet Draft, draft-ietf-manet-aodv-11.txt, work in progress, [7] J.J. Garia-Luna-Aeves and Cruz Marelo Spohn. Soure-Tree Routing in Wireless Networks. In Proeedings of the Seventh Annual International Conferene on Network Protools Toronto, Canada, page 273, Ot [8] M. Gerla. Fisheye State Routing Protool () for Ad Ho Networks. In Internet Draft, draft-ietf-manet-aodv-03.txt, work in progress, [9] Z.J. Hass and R Pearlman. Zone Routing Protool for Ad-Ho Networks. In Internet Draft, draft-ietf-manet-zrp-02.txt, work in progress, [10] A. Iwata, C.-C. Chiang, G. Pei, M. Gerla, and T.-W. Chen. Salable Routing Strategies for Ad Ho Wireless Networks. IEEE Journal on Seleted Areas in Communiations, Speial Issue on Ad-Ho Networks, pages , Aug [11] Mario Joa-Ng and I-T Lu. A Peer-to-Peer Zone-based Two-level Link State Routing for Mobile Ad Ho Networks. IEEE Journal on Seleted Areas in Communiations, 17(8), [12] D. Johnson, D. Maltz, and J. Jetheva. The Dynami Soure Routing Protool for Mobile Ad Ho Networks. In Internet Draft, draft-ietf-manetdsr-07.txt, work in progress, [13] J. Li, J. Jannotti, D. De Couto, D. Karger, and R. Morris. A salable loation servie for geographi ad-ho routing. In Proeedings of the 6th ACM International Conferene on Mobile Computing and Networking (MobiCom 00), pages , August [14] Shree Murthy and J. J. Garia-Luna-Aeves. A routing protool for paket radio networks. In Mobile Computing and Networking, pages 86 95, [15] G. Pei, M. Gerla, X. Hong, and C. Chiang. A Wireless Hierarhial Routing Protool With Group Mobility. In Pro. Wireless Communiations and Networking, New Orleans, [16] C.E. Perkins and T.J. Watson. Highly Dynami Destination Sequened Distane etor Routing (DSD) for Mobile Computers. In ACM SIG- COMM 94 Conferene on Communiations Arhitetures, London, UK, [17] I. Stojmenovi. Position based routing in ad ho networks. IEEE Commmuniations Magazine, pages , July [18] C. Toh. A novel distributed routing protool to support ad-ho mobile omputing. In IEEE 15th Annual International Pheonix Conf. pp , 1996.