Location Prediction Based Routing Protocol for Mobile Ad Hoc Networks

Location Prediction Based Routing Protocol for Mobile Ad hoc Networks Natarajan Meghanathan Department of Computer Science Jackson State University Jackson, MS 39217, USA Email: [email protected] Abstract— We propose a new mobile ad hoc network (MANET) routing protocol called “Location Prediction Based Routing” (LPBR) to simultaneously minimize the number of route discoveries and hop count of the paths for a source-destination session. During a regular flooding-based route discovery, LPBR collects the location and mobility information of nodes in the network and stores the collected information at the destination node of the route search process. When the minimum-hop route discovered through the flooding-based route discovery fails, the destination node attempts to predict the current location of each node using the location and mobility information collected during the latest flooding-based route discovery. A minimum hop Dijkstra algorithm is run on the locally predicted global topology. If the predicted minimum hop route exists in reality, no expensive flooding-based route discovery is needed and the source continues to send data packets on the discovered route; otherwise, the source initiates another flooding-based route discovery. Our ns-2 simulations indicate that LPBR incurs a significantly reduced number of flooding-based route discovery attempts compared with the minimum-hop based and stable path routing protocols. At the same time, the average hop count per path is close to that of the minimum-hop based routing protocols. Key Words: Location prediction, Routing, Ad hoc networks, Route discoveries, Hop count

I.

INTRODUCTION

A mobile ad hoc network (MANET) is a dynamic distributed system of wireless nodes that move independently of each other. MANET routing protocols are either proactive or reactive in nature. Proactive routing protocols determine and maintain routes between any pair of nodes irrespective of their requirement. The reactive on-demand routing protocols determine a route only when required. As the network topology changes dynamically, reactive routing has been preferred over proactive routing [1]. We will focus only on the reactive ondemand routing protocols in this paper. Based on route selection principles, MANET routing protocols can be classified as minimum-weight based and stability-based [2]. Most of the minimum-weight based protocols aim to minimize the number of hops in a path. Stability-based protocols aim for routes with longer lifetimes in order to reduce the number of route discoveries. Performance comparison studies [3] reveal that the stable path protocols could incur as low as half the number of route discoveries incurred by the minimum-hop based protocols, but the average

hop count of stable paths could be as large as twice the minimum hop count. Frequent flooding-based route discoveries incurred by the minimum-hop based protocols significantly consume network bandwidth and congest the network. Stable paths with larger hop count also consume more network bandwidth and reduce frequency reuse. Moreover, nodes that are part of a stable path are used more predominantly compared to the other nodes in the network. Here, we propose a new MANET routing protocol called “Location Prediction Based Routing” (LPBR) protocol that simultaneously minimizes the number of route discoveries as well as the hop count of paths used for a source-destination session. We assume all the nodes are position-aware using techniques like Global Positioning Systems [4] and the clocks across all nodes are synchronized. LPBR works as follows: Whenever a source node has data packets to send to a destination node but does not have a route to that node, it initiates a flooding-based route discovery by broadcasting a Route-Request (RREQ) packet. During this flooding process, each node forwards the RREQ packet exactly once after incorporating its location update vector (LUV) in the RREQ packet. The LUV of a node comprises the node id, the current X and Y co-ordinates of the nodes, the current velocity and angle of movement with respect to the X-axis. The destination node collects the LUV information of all the nodes in the network from the RREQ packets received through several paths and sends a Route-Reply (RREP) packet to the source on the minimum hop path traversed by a RREQ packet. The source starts sending the data packets on the path learnt (based on the RREP packet) and informs the destination about the time of next packet dispatch through the header of the data packet currently being sent. If an intermediate node could not forward a data packet, it sends a Route-Error packet to the source node, which then waits a little while for the destination to inform it of a new route predicted using the LUVs gathered from the latest flooding-based route discovery. If the destination does not receive the data packet within the expected time, it locally constructs the current global topology by predicting the locations of the nodes. Each node is assumed to be currently moving in the same direction and speed as mentioned in its latest LUV. If there is at least one path in the predicted global topology, the destination node sends the source a LPBR-RREP packet on the minimum hop path in the predicted topology. If the predicted path actually exists in reality, the intermediate nodes on the predicted route manage to forward the LPBR-RREP packet to the source. The source uses

This research was supported by the Center for University Scholars at Jackson State University through a research grant for Summer 2008

978-1-4244-2324-8/08/$25.00 © 2008 IEEE. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

the route learnt through the latest LPBR-RREP packet to send the data packets. A costly flooding-based route discovery has been thus avoided. If an intermediate node could not forward the LPBR-RREP packet (i.e., the predicted path did not exist in reality), the intermediate node sends a LPBR-RREP-ERROR packet to the destination informing it of the failure to forward the LPBR-RREP packet. The destination discards all the LUVs and the source initiates the next flooding-based route discovery after timing out for the LPBR-RREP packet. We implemented LPBR in ns-2 [5] and compared its performance with that of the minimum-hop based Dynamic Source Routing (DSR) [6] and the stability based Associativity-Based Routing (ABR) [7], Flow Oriented Routing Protocol (FORP) [8] and Route-lifetime Assessment Based Routing (RABR) [9] protocols. We find LPBR to be significantly successful in reducing the number of route discoveries compared to both the minimum hop based DSR and stable path based ABR, FORP and RABR protocols. Even though, there is some overhead in including the LUV information in the RREQ packets and the next packet sending time information in the data packets, we find this to be a small and useful overhead compared to the huge overhead incurred when using the minimum-hop based DSR (that incurs significantly larger number of route discoveries) and the stable path based routing protocols like ABR, FORP and RABR (that incur more hops per path). Also, as LPBR always opts for the minimum hop paths in the predicted global topology, the average hop count per path for LPBR routes is significantly smaller than that of the stable path routing protocols and is almost close to that of the minimum-hop based DSR. Thus, LPBR reduces the number of route discoveries as well as incurs lower hop count per path. The rest of the paper is organized as follows: In Section II, we present in detail our proposed Location Prediction Based Routing (LPBR) protocol and also explain the different control packets and data packet used in the protocol. Section III presents the simulation conditions used to compare the performance of LPBR with the DSR, ABR, FORP and RABR protocols; explain the simulation results observed with the different routing protocols and highlight the reduction in the number of route discoveries and the average hop count per path achieved with LPBR. Section IV draws the conclusions. II.

LOCATION PREDICTION BASED ROUTING (LPBR) PROTOCOL

LPBR does not require the periodic broadcast of beacons in the neighborhood. It assumes nodes are position-aware and the clocks across all nodes are synchronized. A. Route Discovery to Collect Location Update Vectors When the source has a data packet to send to a destination and is not aware of any route to that node, the source initiates a flooding-based route discovery by sending a Route-Request packet (RREQ) to its neighbors. The source maintains a monotonically increasing sequence number for the floodingbased route discoveries it initiates to reach the destination. Each node on receiving the first RREQ packet (with a sequence number greater than those seen before), will include

its location update vector LUV (comprising the node id, X, Y co-ordinate information, current velocity and angle of movement with respect to the X-axis) in the RREQ packet. The intermediate node also appends its node id in the “Route record” field of the RREQ packet. The LUV and the RREQ packet are shown in Figures 1 and 2 respectively.

Figure 1. Location Update Vector Collected from Each Node

Figure 2. Route Request (RREQ) Packet with the LUVs The destination receives several RREQ packets across different paths and selects the minimum hop path among them using the “Route record” field in these RREQ packets. A Route-Reply (RREP) packet (refer Figure 3) is sent on the discovered minimum hop route to the source. All the nodes receiving the RREP packet will update their routing tables to forward incoming data packets (for the source-destination session) to the node that sent the RREP packet. Note that in order to collect the latest location and mobility information of each node through the LUVs, we intentionally do not let any intermediate node respond to the source with a RREP for a RREQ packet. RREQ packets reach the destination through several paths and will gather the LUVs from several nodes in the network. We consider the time of receipt of the RREQ packet as the time of obtaining the LUV of a node as we expect no major link failures to happen during the time lapsed in between a node sending its LUV in the RREQ packet and the RREQ reaching the destination.

Figure 3. Route Reply (RREP) Packet B. Data Packet Transmission and Route Maintenance The source starts sending data packets to the destination on the route learnt through the RREP packet. In addition to the usual sequence number, source and destination fields, the header of the data packet (refer Figure 4) has three specialized fields: the ‘More Packets’ (MP) field, the ‘Current Dispatch Time’ (CDT) field and the ‘Time Left for Next Dispatch’ (TLND) field. The additional overhead associated with these three header fields amount to only 97 bits per data packet.

Figure 4. Structure of the Header of the Data Packet


The CDT field stores the time as the number of milliseconds lapsed since Jan 1, 1970, 12 AM. If the source has more data to send, it sets the MP flag to 1 and the TLND field to be the number of milliseconds since the CDT of the latest data packet sent. If the source has no more data to send, the MP flag is set to 0 and the TLND field is left blank. As we assume synchronized clocks across all nodes, the destination calculates the end-to-end delay for the data packet based on the local time of receipt of the data packet and the CDT field in the header of the data packet. The destination maintains an average of the end-to-end delay per data packet incurred for the path currently being used to communicate with the source and updates it based on the end-to-end delay suffered by the data packet currently received. If the source node has set the MP flag, the destination node computes the ‘Next Expected Packet Arrival Time’ (NEPAT) as CDT + TLND + 2*Average end-to-end delay per data packet. A timer is started for the NEPAT value. If a link failure occurs due to the two nodes constituting the link drifting away, the upstream node of the broken link informs the source through a Route-Error packet (refer Figure 5). The source on learning the route failure stops sending data packets and waits for the destination to inform it of any new route through a LPBR-RREP packet.

Figure 5. Structure of the Route-Error Packet C. Predicting Node Location using Location Update Vector If the destination does not receive the data packet within the NEPAT time, it will attempt to locally construct the global topology using the location and mobility information of the nodes learnt from the latest flooding-based route discovery. Each node is assumed to be continuing to move in the same direction with the same speed as mentioned in its latest LUV. Let (XuSTIME, YuSTIME) be the X and Y co-ordinates of node u learnt from its LUV collected at time STIME. Let AngleuSTIME and VelocityuSTIME represent the angle of movement with respect to the X-axis and the velocity at which node u is moving. We determine the location of node u at time CTIME, denoted by (XuCTIME, YuCTIME), as follows: Distance traveled by node u from time STIME to CTIME is: DistanceuSTIME-CTIME = (CTIME – STIME + 1)* VelocityuSTIME. Then, XuCTIME = XuSTIME + Offset-XuCTIME and YuCTIME = YuSTIME + Offset-YuCTIME. The offsets in the X and Y-axes depend on the angle of movement and the distance traveled. Offset-XuCTIME = DistanceuSTIME-CTIME * cos(AngleuSTIME) Offset-YuCTIME = DistanceuSTIME-CTIME * sin(AngleuSTIME) where 0˚ ≤ AngleuSTIME ≤ 360˚ Let the network boundaries be given by [0, 0], [Xmax, 0], [Xmax, Ymax] and [0, Ymax]. XuCTIME and YuCTIME are bounded by the constraints: 0 ≤ XuCTIME ≤ Xmax and 0 ≤ YuCTIME ≤ Ymax. If a predicted X and/or Y co-ordinate value falls outside this window, the value is reset to the nearest co-ordinate limit.

D. Path Prediction and Source Notification Based on the predicted locations of each node in the network at time CTIME, the destination node locally constructs the global topology. Note that there exists an edge between two nodes in the locally constructed global topology if the predicted distance between the two nodes (computed based on their predicted locations) is less than or equal to the transmission range of the nodes. The destination node d then locally runs the Dijkstra’s minimum hop path algorithm [10] (with the starting node being the source s) on the predicted global topology. If at least one s-d path exists, the destination sends a LPBR-RREP packet (refer Figure 6) on the minimum hop s-d path with the route information included in the packet.

Figure 6. Structure of the LPBR-RREP Packet Each intermediate node receiving the LPBR-RREP packet updates its routing table to record the incoming interface of the packet as the outgoing interface for any data packet sent from s to d and then forwards the LPBR-RREP packet to the next node on the path to the source node. If the predicted s-d path exists in reality, then the source s is most likely to receive the LPBR-RREP packet before the LPBR-RREP-timer expires. The source now sends the data packets on the route learnt through the latest LPBR-RREP packet received from the destination. The Route-Repair Time (RRT) is the time that lapsed since the source received the Route-Error packet. An average RRT value is maintained at the source as it undergoes several route failures and repairs before the next floodingbased route discovery. The LPBR-RREP-timer (initially set to the route acquisition time) is then set to 1.5*Average RRT value, so that we give sufficient time for the destination to learn about the route failure and generate a new LPBR-RREP packet. Nevertheless, this timer value will be still far less than the route acquisition time that would be incurred if the source were to launch a flooding-based route discovery. Hence, our approach will only increase the network throughput.

Figure 7. Structure of the LPBR-RREP-ERROR Packet E. Handling Prediction Failures If an intermediate node could not successfully forward the LPBR-RREP packet to the next node on the path towards the source, it informs the absence of the route to the destination through a LPBR-RREP-ERROR packet (refer Figure 7). The destination node on receiving the LPBR-RREP-ERROR packet discards all the LUVs and does not generate any new


LPBR-RREP packet. After the LPBR-RREP-timer expires, the source node initiates a new flooding-based route discovery. III.

SIMULATIONS

We use ns-2 (version 2.28) [5] as the simulator for our study. We implemented the LPBR, FORP, ABR and RABR protocols and used the implementation of DSR that comes with ns-2. The MAC layer used is the IEEE 802.11 [11] implementation available in ns-2. The network dimensions used are 650m x 1500m (rectangular network) and 1000m x 1000m (square network). The number of nodes used in each of the above two networks are 50 (low density with an average of 10 neighbors per node) and 100 nodes (high density with an average of 20 neighbors per node). The transmission range of each node is assumed to be 250m. Initially, nodes are uniformly randomly distributed in the network. The performance metrics measured are the average number of route discoveries per s-d session and the average hop count per path, time-averaged over all the s-d sessions. The node mobility model used is the Random Waypoint model [12]. Each node starts moving from an arbitrary location to a randomly selected destination location at a speed uniformly distributed in [0,…, vmax]. Once the destination is reached, the node continues to move by choosing a different target location and a different velocity. The vmax values used are 5, 20 and 50 m/s representing scenarios of low, medium and high node mobility respectively. Traffic sources are constant bit rate (CBR). The number of s-d sessions used is 15. The starting times of these s-d sessions is uniformly distributed between 1 to 50 seconds. Data packets are 512 bytes in size and the packet sending rate is 4 data packets/second. Each data point in Figures 8 through 11 is an average of the data collected using 5 mobility trace files and 5 sets of randomly selected 15 s-d sessions. A. FORP, ABR and RABR In FORP, each node keeps track of the predicted lifetimes of the links with its neighbors. The predicted lifetime of a route is the minimum of the predicted lifetimes of the constituent links of the route. FORP selects the route with the maximum predicted route lifetime. In the case of ABR, each node maintains the number of beacons (called associativity ticks) received from its neighbors. If the number of associativity ticks received from a neighbor exceeds the Associativity Threshold (calculated based on the relative velocity of the nodes in the network), then the link with the neighbor is said to be a strong link; otherwise the link is termed a weak link. ABR chooses the route that has the maximum proportion of strong links. In case of a tie, then ABR chooses the minimum hop path among those paths that have the same maximum proportion of strong links. In the case of RABR, each node keeps track of the affinity of the links with its neighbors by measuring the signal strength of the beacons received from the neighbors. Link affinity is a measure of link stability. If two nodes are approaching each other, then the link between them has a high affinity value. If

two nodes are moving away from each other, then the link between them has a low affinity value. The affinity of a route is the minimum of the affinity values of the constituent links. RABR selects the route with the maximum affinity value. B. Hop Count per Path Figures 8.1 through 8.3 and 9.1 through 9.3 illustrate that the hop count incurred by LPBR routes is almost close to that incurred by DSR routes and is significantly smaller compared to that incurred by the stable path protocols. This indicates the effectiveness of the LPBR route prediction approach. The hop count of the routes predicted upon a route failure is close to being the minimum in the network at that instant of time.

Figure 8.1: vmax = 5m/s

Figure 9.1: vmax = 5 m/s





Figure 8: Hop Count per Path in Square Network

Figure 9: Hop Count per Path in Rectangular Network

In networks of square shape, the nodes are uniformly distributed in both the X and Y directions. There is no significant increase in the probability of the network remaining connected when we increase the number of neighbors per node from 10 to 20. Hence, the hop count of the routes chosen does not depend much on node distribution and node mobility. DSR has the minimum hop paths among all the routing protocols. The hop count of the LPBR routes is larger than that of the DSR routes only by 6 to 12% and is about 0.85 times to that of the FORP routes, 0.7 times to that of the ABR routes and 0.6 times to that of the RABR routes. In networks of rectangular shape, the nodes are more preferentially distributed towards the larger dimension. As a result, the hop count of the routes will be larger compared to that incurred in square networks of the same total area. The magnitude of the difference depends on the routing protocol and the network density. Comparing the hop count incurred by LPBR with those incurred by the other routing protocols, we observe that the hop count of LPBR routes is 2-6% more than that of DSR, 0.55-0.6 times less than that of ABR, 0.5 less than that of RABR and 0.85 times less than that of FORP.








Figure 10: Route Discoveries in Square Network

Figure 11: Route Discoveries in Rectangular Network

C. Number of Route Discoveries Figures 10.1 through 10.3 and 11.1 through 11.3 illustrate that LPBR incurs the least number of route discoveries when compared to the other four routing protocols. The stable path protocols undergo a reduced number of route discoveries compared to the minimum hop based DSR. The instability of minimum hop paths could be attributed to the larger physical distance (on average close to 80% of the transmission range of the nodes) of the constituent hops on the minimum hop path (edge effect [13]). Even though LPBR routes are also prone to edge effect, the effectiveness and accuracy of the location prediction approach helps to avoid the flooding-based route discoveries as much as possible. In networks with square topology, the number of route discoveries is not very much influenced by the “edge effect”. When it comes to ranking the routing protocols in terms of the magnitude of the number of flooding-based route discoveries, we observe LPBR incurs the minimum number of route discoveries and DSR incurs the maximum. In networks of rectangular topology, the routes are influenced by the “edge effect”. For a given node mobility and network density, the probability of a route failure is more in a rectangular topology compared to that in a square topology because, the average hop count of a route incurred by any routing protocol in a rectangular topology is greater than that incurred with a square topology when run under identical conditions. Nevertheless, all of the above observations in the simulation results indicate that LPBR incurs the lowest number of route discoveries among all the five routing protocols. IV.

CONCLUSIONS

protocol that simultaneously minimizes the number of route discoveries as well as the hop count per path. We ran extensive simulations by varying the network density, node mobility and network topology shape. We compared the performance of LPBR with that of the minimum-hop based DSR and the stable path routing protocols like FORP, ABR and RABR. LPBR incurs the least number of route discoveries and the hop count per path is only at most 12% more than that incurred with the minimum-hop DSR. The number of route discoveries is often significantly low compared to that incurred with DSR, ABR and is about 60 to 80% of the number of route discoveries incurred by FORP and RABR for most of the simulation conditions. This indicates the effectiveness of the location prediction approach in LPBR. As there exists no single routing protocol that simultaneously minimizes the number of route discoveries as well as the hop count per path, LPBR is a valuable addition to the MANET literature. REFERENCES [1]

[2]

[3]

[4] [5] [6]

[7] [8]

[9]

[10] [11]

[12]

[13]

J. Broch, D. A. Maltz, D. B. Johnson, Y. C. Hu, J. Jetcheva, “A Performance Comparison of Multi-hop Wireless Ad Hoc Network Routing Protocols,” Proceedings of the 4th ACM Mobile Computing and Networking Conference, pp.85–97, Oct. 1998. N. Meghanathan and A. Farago, “Survey and Taxonomy of 55 Unicast Routing Protocols for Mobile Ad Hoc Networks,” Technical Report UTDCS-40-04, University of Texas at Dallas, Nov. 2004. N. Meghanathan, “Stability-Energy Consumption Tradeoff among Mobile Ad hoc Network Routing Protocols,” Proceedings of the 3rd International Conference on Wireless and Mobile Communications, Mar. 2007. B. Hofmann-Wellenhof, H. Lichtenegger and J. Collins, Global Positioning System: Theory and Practice, 5th ed., Springer, Sept. 2004. L. Breslau et. al, “Advances in Network Simulation,” IEEE Computer, vol. 33, no. 5, pp. 59 – 67, May 2000. D. B. Johnson, D. A. Maltz and J. Broch, “DSR: The Dynamic Source Routing Protocol for Multi-hop Wireless Ad hoc Networks,” in Ad hoc Networking, Chapter 5, C. E. Perkins, Eds. Addison Wesley, pp. 139 – 172, 2000. C-K. Toh, “Associativity-Based Routing for Ad hoc Mobile Networks,” IEEE Personal Communications, vol. 4, no. 2, pp. 103 – 139, Mar. 1997. W. Su, S-J Lee and M. Gerla, “Mobility Prediction and Routing in Ad Hoc Wireless Networks,” International Journal of Network Management, vol. 11, no. 1, pp. 3-30, Jan.-Feb. 2001. S. Agarwal, A. Ahuja, J. P. Singh and R. Shorey, “Route-Life Time Assessment Based Routing Protocol for Mobile Ad Hoc Networks,” Proceedings of IEEE International Conference on Communications, pp. 1697 – 1701, Jun. 2000. T. H. Cormen, C. E. Leiserson, R. L. Rivest and C. Stein, “Introduction to Algorithms,” 2nd Edition, MIT Press/ McGraw-Hill, Sept. 2001. G. Bianchi, “Performance Analysis of the IEEE 802.11 Distributed Coordination Function,” IEEE Journal of Selected Areas in Communication, vol. 18, no. 3, pp. 535-547, Mar. 2000. C. Bettstetter, H. Hartenstein and X. Perez-Costa, “Stochastic Properties of the Random-Way Point Mobility Model,” Wireless Networks, pp. 555 – 567, vol. 10, no. 5, Sept. 2004. G. Lim, K. Shin, S. Lee, H. Yoon and J. S. Ma, “Link Stability and Route Lifetime in Ad hoc Wireless Networks,” Proceedings of the International Conference on Parallel Processing Workshops, pp. 116– 123, Aug. 2002.

The high-level contribution of this paper is the proposal and development of a new mobile ad hoc network routing protocol called Location Prediction Based Routing (LPBR)
