Dynamic Route Switching Protocol in Mobile Ad Hoc Networks

Dynamic Route Switching Protocol in Mobile Ad Hoc Networks Abstract Due to limited bandwidth, how to decrease routing overhead is an important issue in mobile ad hoc networks. Several multipath routing protocols have been proposed to reduce the frequency of route discovery. However, in rapidly changed network topologies, some backup paths may be broken before a host uses them. Additionally, to obtain multiple loop-free paths, some protocols only select backup paths that are equal or shorter than the first found path. Longer routing paths are discarded even if they can stay available longer than the primary path. This paper presents a dynamic route switching protocol (DRSP) to locate more useful and long-lived routing paths. The protocol not only considers node mobility but energy consumption. Based on the prediction, invalid backup paths are erased automatically so the possibility of unsuccessful delivery can be reduced. The mobility prediction also can adjust appropriate transmission power. The DRSP protocol was implemented on ns-2 and simulation results show that the DRSP outperformed AODV, DSR, and AOMDV. Keywords: mobile ad hoc networks, routing protocol, multiple paths, power failure.

1 Introduction A mobile ad hoc network is one type of wireless network that has no assistance of communication infrastructures, such as wireless access points and base stations. Instead of communicating via a centralized infrastructure, each host acts as a router to forward packets for other nodes. When a source node sends data packets to a destination node that is not within the source s transmission range, the packets must be forwarded by its neighbors. The neighbors forward the packets to the destination hop by hop. Therefore, it is important to establish routes between hosts in mobile ad hoc networks. Previous routing protocols for ad hoc networks can be roughly categorized as table-driven and on-demand. Destination Sequenced Distance Vector routing protocol (DSDV), one of the table-driven protocols, was based on Bellman-Ford routing algorithm [1]. Each host maintains all possible routes in its routing table. Each node needs to broadcast routing table advertisements in a period of time. When the topology of networks is changed, each node has to forward update information to maintain table consistency. The frequent updates may cause network congestion. With on-demand routing protocols, such as Dynamic Source Routing (DSR) protocol, Ad Hoc On-demand Distance Vector routing (AODV) protocol, source nodes build routes only when they need to communicate with destination nodes [2] [3]. The on-demand scheme can reduce the control overhead compared to the table-driven approach. Frequent route discovery in dynamic networks arises the routing overhead and end-to-end delay. Several multipath protocols have been proposed to alleviate routing overhead and reduce transmission time. These protocols maintain several potential paths between hosts that are found in the route discovery phase. Previous multipath protocols have several drawbacks. First, a host selects a primary path to forward packets. When the primary path is broken, it uses a backup path for transmission. However, backup paths may be broken before a host uses them to forward packets. Switching to a backup path may cause more packet loss. Besides, some protocols only maintain backup paths that are equal or shorter than the primary path to avoid routing loops. The scheme forbids a host accepting longer paths even if they will stay available longer than the primary path. For example, the route A-C-D-E-B will be discarded if the route A-F-B is found first (see Figure 1). In addition, power consumption is an essential 1

topic in mobile ad hoc networks. The transmission power is typically fixed during transmission. When a receiver moves closer to a sender, the used transmission power will be more than needed. The transmission power should be adjusted based on the distance between two nodes. This paper presents a dynamic route switching protocol (DRSP) for mobile ad hoc networks. A mobility prediction method is adopted to calculate route expiration time (RET). Based on the information, a loop-free route update scheme is used to obtain backup paths that are longer-lived than the primary path. Each mobile node measures its battery expiration time at regular intervals. If the node expiration time is lower than the threshold, a source node will not select routes containing the node in route discovery phase. A route maintenance mechanism removes stall backup paths according to the route expiration time. A host that is going to run out of battery forwards route error packets to notify its upstream nodes. The source nodes thus can eliminate invalid paths immediately before they are broken. With the DRSP, the distance between two nodes can be estimated so a host can use appropriate transmission power to deliver packets. 2 Related Work On-demand routing protocols do not maintain or constantly update routing tables for the latest network topology. Both DSR and AODV are popular on-demand routing protocols [2] [3]. They typically contain two major processes: route discovery and route maintenance. When a host wishes to establish a route, it will initiate a route discovery process. Route request (RREQ) packets issued by the source Figure 1: A drawback of loop-free scheme. 2

are flooded to the network. Intermediate nodes receiving the RREQs build several reverse paths to the source. When the destination receives an RREQ, it generates a route reply (RREP) packet and sends the packet via the reverse paths. As the RREPs are received by intermediate nodes, forward paths to the destination are established. After the route discovery process, the source find out all possible routes and it will determine a route to forward data packets. When a link failure is detected, a route maintenance process will be initiated. The host sends route error (RERR) packets to its upstream nodes, and then they forward the packets back to all source nodes using the broken link. The source receiving an RERR removes the broken route. If the source still needs to send packets, another route discovery will be proceeded. AODV-BR, a protocol based on AODV, establishes multiple paths without transmitting any extra routing packet [4]. The primary path and backup paths form a mesh structure that is similar to a fish bone. AOMDV is an extension of AODV that has multiple loop-free and link-disjoint paths [5]. In Split Multipath Routing protocol (SMR), an extension of DSR, there are two routes for a destination in a host s routing table [6]. Data packets are split into the two paths so some network congestion can be avoided. To obtain multiple loop-free paths, AOMDV and SMR only accept backup paths that are equal or shorter than the primary path. However, the first found path is typically the shortest so some useful (loop-free) paths are discarded. In mobile ad hoc networks, determining the stability of links is an essential issue for improving routing performance. A mobility prediction mechanism uses location information to calculate the disconnection time of routes [7]. Location information provided by the GPS is piggybacked in routing packets. With coordinates, velocity, movement direction of two nodes, the link expiration time (LET) can be computed. A preemptive routing protocol uses the signal power strength to predict the link stability [8]. When a node moves into the preemptive region, a warning packet is sent to upstream nodes for route discovery. There are also several methods using probability models for determining the link availability [9] [10]. Several energy-aware routing protocols were proposed to consume power efficiently and maximize 3

lifetime for mobile hosts. The essence of the protocols is to distribute power consumption evenly. Minimum Total Transmission Power Routing (MTPR) is one of the protocols that use a formula to derive total transmission power for routes [11]. With the information, a source can obtain a route with minimum total transmission power from all possible routes. Minimum Battery Cost Routing (MBCR) utilizes the remaining battery capacity of each host to select routes [12]. Min-Max Battery Costing Routing (MMBCR) considers that the node s remaining battery is greater than a threshold [13]. The Minimum Drain Rate (MDR) uses both energy drain rate and remaining battery power to determine routes for communication [14]. In order to reduce the transmission power, Power Control Routing (PCR) divides transmission power into N levels [15]. A host can choose suitable transmission power for sending packets. A power control protocol can also be implemented in MAC layer [16]. Moreover, the routing overhead can be distributed among mobile nodes by adopting a load balancing scheme [17]. 3 Dynamic Route Switching Protocol Dynamic route switching protocol (DRSP) is both location-aware and power-aware. A mobility prediction method is used to predict the route expiration time (RET) [7]. Power information is utilized to measure the node expiration time. A power control method is also designed to reduce transmission energy. 3.1 Prediction for Power Expiration Time The power expiration time can be estimated by the future power drain rate and the remaining battery capacity. The future power drain rate could be predicted as an exponential average of the previous power drain rate. D t+1 is the predicted value for future power drain rate and D t represents the previous average power drain rate from beginning to t th second. In order to save memory space, a host only stores the information from (t j) th to t th seconds. The corresponding prediction function can be defined as: D t+1 = αd t + (1 α)αd t 1 +... + (1 α) j αd t j (1) 4

In the formula, D i can be derived like this: D i = C 0 C i t i (2) C i is the remaining battery capacity at i th second, C 0 is the initial battery capacity, and t i represents the interval from beginning to i th sec. In formula (1), D t+1 is composed by the recent and past power drain rates, and parameter α is related with weight of the recent and past information. When α becomes larger, the recent history s weight becomes higher. On the contrary, the past history s weight becomes higher. Let C now be the current remaining battery capacity, and the formula can be derived: T expire = C now D t+1 (3) In the DRSP, each host should keep track of its power information periodically. 3.2 Power Control Method In most existing wireless networks, the transmission power is set as a constant. Like Lucent s WaveLAN, the radio propagation range is defined as 250 meters. Nevertheless, the constant transmission power may cause waste of the battery capacity. To save energy consumption, a mobile host should change its transmission power adaptively according to the distance to the receiver. Assume that d is the distance to a neighbor and P D is the needed energy for maximum transmission range D. P adjusted represents the adjusted transmitted signal power and φ is a parameter to increase the predicted distance to neighbor. If (d + φ) is more than D, it will be replaced by D. P adjusted = P D (d + φ) 2 D 2 (4) 5

3.3 Dynamic Route Switching Protocol 3.3.1 Assumptions Each host in the mobile ad hoc networks is equipped with GPS. Location information is piggybacked in routing and data packets. To predict power expiration time, each host should have the ability to read the remaining battery capacity from the power management components. 3.3.2 Data Structure In the DRSP, each mobile host maintains five data structures: a multiple path routing table, a neighbor table, a power information table, a route request table for, and a route reply table. A multiple path routing table contains path sets for each requested destination. Every entry in the table is created and updated when the host receives a routing packet. If there is no entry for the destination in the routing table, it will create a new entry. If the host has a route for the destination, it will check whether the new path conforms to the route update rule. Each mobile host has a timer, called RefreshTimer, for purging stale entries periodically. Each entry in the table contains destination address, next hop address, hop count, sequence number, route expiration time (RET), maximum route expiration time (MRET), pointer to the list of backup paths, and list of the previous nodes for the route. The neighbor table is a list containing network addresses and location information about neighbors. If a host receives a route request packet or a route reply packet from an unknown neighbor, it will create a new entry for the neighbor. If the neighbor exists in the neighbor table, its location information will be updated. Each entry of the neighbor table includes neighbor address, neighbor location information, and the latest time to create or update the table. The power information table contains a list of previous power drain rates. A PowerRefreshTimer is utilized to purge the power consumption information. Each entry in the table contains the initial energy and average power drain rates. The route request table stores a list of source addresses and broadcast IDs. The pair (source address, broadcast ID) identifies a unique route request (RREQ) packet. A host increases its broadcast ID after a new RREQ is sent. If a host receives an RREQ with a new pair (source address, broadcast ID), it will add the pair to the table. Therefore, the 6

route request table is able to determine whether the RREQ is redundant or not. The route reply table includes three entries: destination address, source address, and destination sequence number. The tuple (destination address, source address, destination sequence number) is used to identify a route request (RREP) packet. 3.3.3 Calculate Multiple Loop-free Paths Duplicate route request (RREQ) packets are discarded in single path on-demand routing protocols. In order to obtain all potential paths, all RREQ copies should be processed. However, it could generate routing paths with loops. The DRSP includes a new route update scheme to avoid routing loops. The scheme allows a host to accept backup paths that are longer than the primary path. The value "maximum route expiration time" (described in Section 3.3.2) is set to zero in the beginning. When a path with longer route expiration time is found, the maximum route expiration time is updated. Two rules are used for determining if the route needs update. First, a host keeps only routes with the highest destination sequence number. Second, for the same sequence number, a host accepts a route if the route has a smaller hop count or a larger route expiration time. A path with loops has a larger hop count than the correspondent loop-free path. Moreover, the value of RET in the RREQ becomes smaller due to the LET decreasing during flooding. Therefore, a route containing loops will not be accepted by the DRSP for the same route request. The route update strategy is described in Figure 2. 3.3.4 Route Discovery In route discovery process, both mobility prediction and power failure prediction are applied to select backup paths. The route discovery process has two major phases: route request phase and route reply phase. The route discovery process will be initiated when a route is requested by a source node and there is no information about the route in its routing table. First, the source node generates an RREQ and then floods the packet to networks. The RREQ includes source address, destination address, hop count, broadcast ID, sequence number, location information, route expiration time, and TTL. The RREQ is propagated to neighbors within the source s transmission range. The neighbors, receiving the RREQ, 7

Definitions: rt1: the route entry to node S rt2: the route entry to node D src seqno: the source sequence number in the routing packet dst seqno: the destination sequence number in the routing packet seqno: the sequence number in the routing table ret: the route expiration time in the routing packet mret: the maximum route expiration time in the routing table rq: a route request packet from the source S to the destination D rp: a route reply packet from the destination D to the source S procedure route update begin For an intermediate node receives rq if (rq.src seqno > rt1.seqno) then erase the multipath list; rt1.mret 0; insert the new path into the multipath list; else if (rq.src seqno = rt1.seqno) if((rq.hopcount < rt1.hopcount) or (rq.ret > rt1.mret)) then insert the new path into the multipath list; endif else drop the packet rq; endif For an intermediate node receives rp if (rp.dst seqno > rt2.seqno) then erase the multipath list; rt2.mret 0; insert the new path into the multipath list; else if (rp.dst seqno = rt2.seqno) if((rp.hopcount < rt2.hopcount) or (rp.ret > rt2.mret)) then insert the new path into the multipath list; endif else drop the packet rp; endif end Figure 2: Algorithm for route update. 8

continue to broadcast the packet. When an intermediate node receives the RREQ, it will perform following procedures (detailed algorithm shown in Figure 3). 1. The node measures its node expiration time (NET) first. If the NET is lower than threshold, the host will discard the RREQ. The node will run out of battery soon so the routes should not contain the node. 2. The node determines whether the RREQ is redundant or not by checking the tuple (source address, broadcast ID). If the RREQ is not redundant, the node will refresh the location information of the RREQ and forward the packet to neighboring nodes. On the other hand, it will drop the RREQ. 3. The node decreases TTL of the RREQ by one. If the TTL is smaller than zero, the host will drop the RREQ. 4. The node calculates the link expiration time (LET) to the previous node. If the LET is smaller than the route expiration time (RET) stored in the RREQ, it will replace the RET by LET. 5. To transmit route reply packets to the source, the node builds a reverse path to the source based on the route update rule mentioned in Section 3.3.3. When the destination receives the route request packet, it will send route reply (RREP) packet to the source along the reverse paths created previously. The RREP contains source address, destination address, hop count, destination sequence number, location information, and RET. The RREPs is delivered to the source through the reverse paths. Each intermediate node, forwarding the RREP, performs following operations: 1. The node determines whether the RREP is redundant or not by checking the tuple (destination address, source address, and destination sequence number). If the RREP is not redundant, the node will refresh the location information of the RREP. On the other hand, it will drop the RREP. 2. The node calculates the LET for the sender based on the RREP s location information. If the LET is smaller than the RET stored in the RREP, the LET will replace the RET. 9

Definitions: rt: the route entry to the source S seqno: the sequence number in the routing table rt src: the source IP or node ID in the routing packet src seqno: the source sequence number in the routing packet dst: the source IP or node ID in the routing packet dst seqno: the destination sequence number in the routing packet hop: the hop count in the routing packet in / pn: the IP or node ID of the intermediate node / previous node bid: the broadcast ID RET: the route expiration time in the routing packet nb: the neighbor table entry to the previous node rq / rp: a route request packet / a route reply packet procedure Recv RouteRequest begin if(in = rq.src) then drop the packet rq; else compute NET = the node expiration time of this node; TTL = TTL - 1; if((net < threshold) or (bid lookup(rq.src, rq.bid) NULL) or (TTL < 0) ) then drop the packet rq; else rt.update; bid insert(rq.src,rq.bid); nb update(pn, location information); compute LET = the link expiration time to pn; if((rq.ret = INFINITY LET INFINITY) or (rq.ret INFINITY LET INFINITY rq.ret > LET)) then rq.ret LET; endif endif endif if(in = rp.dst) then rp.dst seqno = max(seqno, rq.dst seqno) send a route reply packet to rq.src; else rq.hop = rq.hop + 1; replace the rp s location information by the node s information; relay the packet rq; endif end Figure 3: Algorithm for route request. 10

3. If the path conforms to the route update rule, the node will insert the path to its forward path list. If not, the node will ignore the path and discard the RREP. 4. The node creates or updates the entry of neighbor table based on the location information and timestamp. 5. The node forwards the RREP to next nodes of reverse paths. After the above process, multiple routing paths are obtained. The DRSP selects the shortest one as the primary path. If there are more than one shortest paths, a source will select the path found first. When the primary path is broken, a host switches data traffics to the shortest backup path. During data transmitting, a host measures the distance to the next hop using the location information stored in its neighbor table. The host can adjust the transmission power dynamically. Additionally, to improve the availability of backup paths, a host periodically removes the stale backup paths based on the predicted lifetime. 3.3.5 Route Maintenance Strategy Link failures in the ad hoc networks are caused by mobility, congestion, packet collisions, node failures, and so on. In the DRSP, the link layer feedback from IEEE 802.11 is utilized to detect link failures. If a node sends packets along the broken link, it will receive a link layer feedback. When a node detects a link break, it broadcasts a route error (RERR) packet with an unreachable destination array to its neighbors. After a node receives the RERR, it will remove every entry in its routing table that uses the broken link, and then rebroadcast the RERR. On the other hand, if the node does not use the broken link, it just discards the packet. Unlike AODV, the route error packets contain not only about the primary path but also the backup paths. When the source node receives the RERRs, it removes all broken routing entries and uses the shortest available backup paths as its primary path. The source node will initiate a route discovery process if all backup paths are broken. A link failure can be predicted by another way with the DRSP. When a host receives a data packet from its neighbor, it computes the LET of the link. If the host finds that the value of LET is lower 11

than threshold, the host will send RERRs to the source nodes because the link between the host and its neighbor will break soon. Additionally, each host calculates its power expiration time periodically. When the value of node expiration time is lower than threshold, the host will send RERRs to notify its neighbors. Therefore, a host can remove the invalid backup paths preemptively and reduce packet loss. 4 Performance Evaluation 4.1 Simulation Environment The DRSP was evaluated using the ns-2 simulator version 2.1b9a [18] with the CMU s multi-hop wireless extensions. The IEEE 802.11 distributed coordination function (DCF) was used as the medium access control protocol. The physical radio characteristics were based on Lucent s WaveLAN. Wave- LAN was direct spread spectrum radio and the channel had radio propagation range of 250 meters and capacity of 2 Mb/sec. The DSR, AODV, AOMDV, and DRSP were compared in the simulation. The simulation model was consisted of 50 mobile nodes randomly distributed in an 1500*300 rectangular area. The traffic pattern consists of 30 constant bit rate (CBR) sources sending 512-byte packets at a constant rate 4 packets per second. The random waypoint model was used for node movement. Each node selected a random destination and moved to the position with a specified speed. When a node arrived the destination, it stopped for a predefined pause time. In the simulation, the pause time was modelled as normal distribution and the mean value was 60 seconds. Besides, the velocity of each node was also normal distributed. The movement patterns were generated by using 6 different average velocities: 0, 2.5, 5, 10, 15, 20 m/s. The total simulation time was 900 seconds. 4.2 Simulation Results The following metrics were used to evaluate the routing protocols: Average number of paths: The average number of paths found in each route discovery process. 12

Successful rate of backup paths: This metric is defined as a probability of successful switching data to backup paths. Packet delivery ratio: The number of data packets delivered to the destinations to the number of data packets sent by the sources. Normalized routing load: The number of routing packets transmitted per data packets delivered. End-to-end delay: Average time between data packets received by the destinations and data packets sent by CBR sources. The data were collected only for successfully delivered packets. Throughput: The total size of data packets that are received in CBR destinations per second. Normalized energy consumption: The value of energy consumption per data packets delivered. 4.2.1 Without Node Failures Table 1: Average Number of Paths and Successful Rate for Backup Paths Mean speed 2.5 (m/s) 5 (m/s) 10 (m/s) 15 (m/s) 20 (m/s) Metric Number Rate Number Rate Number Rate Number Rate Number Rate AOMDV 2.168 0.762 2.119 0.700 2.155 0.589 2.148 0.529 2.056 0.490 DRSP 2.393 0.926 2.351 0.909 2.446 0.848 2.426 0.810 2.367 0.752 Power failure was not considered in the first set of simulations. Every node functioned correctly during the simulation. Table 1 shows the average number of paths and the successful rates of backup paths for both AOMDV and DRSP. The DRSP had more routing paths in each situation. Moreover, the successful rates of backup paths in DRSP were higher than AOMDV in each scenario because DRSP removed stale backup paths periodically. The results proved that DRSP provided more useful paths than AOMDV. Figure 4 compares the packet delivery ratios for the four protocols. Though AOMDV provided backup routing paths, the packet delivery ratio of AOMDV was lower than AODV and DSR. AOMDV did not update available backup routing paths so the hosts may switch data traffics to broken 13

routes. With DRSP, the mobility prediction method was used to build backup paths and the dynamic route maintenance also helped to erase invalid backup routes. Therefore, the packet delivery rates with DRSP were higher than AOMDV. Both multipath protocols had smaller normalized routing loads (shown in Figure 5). The DRSP provided more useful backup paths so its normalized routing load was better than AOMDV. Figure 6 shows the advantage of multipath routing protocols for reducing end-to-end delay. Instead of initiating a new route discovery, using backup paths eliminated the overhead and delay of the route discovery process. The DRSP performed better than other protocols due to more reliable backup paths. The Packet Delivery Rate (%) 100 95 90 85 80 AODV AOMDV DSR DSRP 75 0 5 10 15 20 Mean Speed (m/s) Figure 4: Packet delivery ratio. Normalized Routing Load (packets) 18 16 14 12 10 8 AODV 6 AOMDV 4 DSR 2 DRSP 0 0 5 10 15 20 Mean Speed (m/s) Figure 5: Normalized routing load. 14

AOMDV did not perform stably and had larger delay than AODV and DSR when moving speed was low. Broken backup paths led to local repair process for building a new route from the intermediate nodes to the destinations so the end-to-end delay for AOMDV was affected. Figure 7 is a comparison for the throughput of the protocols. Due to smaller number of routing packets and the lower end-toend delay, the DRSP transmitted more data packets than other protocols during the simulation time. Figure 8 illustrates that DRSP consumed energy more effectively than AODV, DSR, and AOMDV. Though AOMDV can reduce the number of routing packets, it only performed a little better than AODV and DSR. The DRSP reduced the transmission energy by the power control method and had fewer 0.09 0.08 0.07 End-to-end Delay (s) 0.06 0.05 0.04 0.03 0.02 0.01 AODV AOMDV DSR DRSP 0 0 5 10 15 20 Mean Speed (m/s) Figure 6: End-to-end delay. Figure 7: Throughput. 15

routing packets so the normalized energy consumption was further enhanced. 4.2.2 With Node Failures Table 2: Linear Power Consumption Model of Lucent IEEE 802.11 WaveLAN PC Card Packet type Energy consumption (µw) broadcast send 1.9 PacketSize + 250 point-to-point send 1.9 PacketSize + 420 broadcast receive 0.50 PacketSize + 56 broadcast send 0.42 PacketSize + 330 The initial energy for each node was modelled using normal distribution and the mean value was 30 joules. Based on the Lucent IEEE 802.11 WaveLAN s specification, the linear power consumption model coefficients for data sending and receiving are shown in Table 2. In the end of the simulation, about ten nodes produced power failure. In Table 3, the average number of paths for each protocol was smaller than the results without node failures. Node failures decreased the density of mobile hosts so less nodes could help to establish multiple routing paths. However, the successful rates of DRSP were almost the same compared to prior results. In Figure 9, the packet delivery ratio for AODV, DSR, and AOMDV decreased compared Normalized Energy Consumption (joules) AODV AOMDV DSR DRSP 0 0 5 10 15 20 Mean Speed (m/s) Figure 8: Normalized energy consumption. 16

to the previous results. On the other hand, due to node failure prediction method and dynamic route maintenance, the packet delivery ratio of DRSP was better than other protocols. The performance of the DRSP was not much affected by the faulty nodes. Table 3: Average Number of Paths and Successful Rate for Backup Paths Mean speed 2.5 (m/s) 5 (m/s) 10 (m/s) 15 (m/s) 20 (m/s) Metric Number Rate Number Rate Number Rate Number Rate Number Rate AOMDV 1.742 0.731 1.829 0.882 1.794 0.571 1.913 0.514 1.845 0.486 DRSP 1.943 0.919 2.036 0.904 1.973 0.852 2.113 0.807 2.104 0.763 100 Packet Delivery Rate (%) 95 90 85 80 AODV AOMDV DSR DRSP 75 0 5 10 15 20 Mean Speed (m/s) Figure 9: Packet delivery ratio. 20 Normalized Routing Load (packets) 18 16 14 12 10 8 AODV 6 AOMDV 4 DSR 2 DRSP 0 0 5 10 15 20 Mean Speed (m/s) Figure 10: Normalized routing load. 17

The AOMDV had better routing load than AODV and DSR (see Figure 10). With node failure prediction, DRSP outperformed other protocols in all varying speeds. A source node could avoid selecting a path with nodes that were going to run out of energy. In Figure 11, DRSP had the best end-to-end delay with all different speed settings because some local repair processes were avoided. AOMDV did not always perform better than both AODV and DSR due to some useless backup routing paths. Figure 12 illustrates that DRSP had the best throughput among the protocols. As shown in Figure 13, the DRSP saved more energy than other protocols. 0.09 0.08 0.07 End-to-end Delay (s) 0.06 0.05 0.04 0.03 AODV 0.02 AOMDV DSR 0.01 DRSP 0 0 5 10 15 20 Mean Speed (m/s) Figure 11: End-to-end delay. Figure 12: Throughput. 18

5 Conclusion The DRSP was developed for providing more useful backup paths for data transmission in mobile ad hoc networks. The DRSP utilizes both mobility prediction and power failure prediction to select longer-lived backup paths. The dynamic route maintenance scheme also removes invalid backup routes at regular intervals. Therefore, the better reliability of the backup paths improves routing performance. Moreover, the DRSP uses power control method to reduce required transmission energy. Simulation results showed that the DRSP can locate more effective backup paths than the AOMDV. The successful delivery rates of the DRSP were about 30% higher than AOMDV. The routing load of the DRSP was also smaller than AODV, DSR, and AOMDV. For the end-to-end delay, the DRSP performed 30% faster than AODV and DSR, and 20% faster than AOMDV. Due to the less number of routing packets and the smaller end-to-end delay, the DRSP achieved the best throughput. The DRSP also reduced 20% energy consumption compared to other routing protocols. References [1] C. E. Perkins and P. Bhagwat, Highly Dynamic Destination-Sequenced Distance Vector Routing for Mobile Computers, Proceedings of ACM SIGCOMM, pp. 234 244, Dec. 1994. [2] D. B. Johnson and D. A. Maltz, Dynamic Source Routing in Ad Hoc Wireless Networks, Mobile Computing, vol. 353, pp. 153 179, 1996. 0.6 Normalized Energy Consumption (joules) 0.5 0.4 0.3 0.2 AODV AOMDV 0.1 DSR DRSP 0 0 5 10 15 20 Mean Speed (m/s) Figure 13: Normalized energy consumption. 19

[3] C. E. Perkins and E. M. Royer, Ad-Hoc On-Demand Distance Vector Routing, Proceedings of IEEE Workshop on Mobile Computing Systems and Applications, pp. 90 100, Feb. 1999. [4] S. J. Lee and M. Gerla, AODV-BR: Backup Routing in Ad Hoc Networks, Wireless Communications and Networking Conference, pp. 1311 1316, Sept. 2000. [5] M. K. Marina and S. R. Das, Ad hoc On-demand Multipath Distance Vector Routing, Review of ACM SIGMOBILE Mobile Computing and Communications, pp. 92 93, July 2002. [6] S. J. Lee and M. Gerla, Split Multipath Routing with Maximally Disjoint Paths in Ad Hoc Networks, Proceedings of the IEEE International Conference on Communications, pp. 3201 3205, June 2001. [7] W. Su, S. J. Lee, and M. Gerla, Mobility Prediction in Wireless Networks, Proceedings of the IEEE International Conference on Military Communications, pp. 491 495, Oct. 2000. [8] T. Goff and N. B. Abu-Ghazaleh, Preemptive Routing in Ad Hoc Networks, Proceedings of the Seventh Annual International Conference on Mobile Computing and Networking, pp. 43 52, July 2001. [9] B. McDonald and T. Znati, A Path Availability Model for Wireless Ad-Hoc Netwotrks, Proceedings of IEEE Conference on Wireless Communications and Networking, pp. 35 40, Sept. 1999. [10] S. Jiang, D. He, and J. Rao, A Prediction-Based Link Availability Estimation for Mobile Ad-Hoc Networks, Proceedings of IEEE INFOCOM, pp. 1745 1752, Apr. 2001. [11] K. Scott and N. Bambos, Routing and Channel Assignment for Low Power Transmission in PCS, Proceedings of IEEE International Conference on Universal Personal Communications, pp. 498 502, Oct. 1996. [12] S. Singh, M. Woo, and C. Raghavendra, Power-aware Routing in Mobile Ad Hoc Networks, Proceedings of Annual International Conference on Mobile Computing and Networking, pp. 181 190, Oct. 1998. [13] C.-K. Toh, H. Cobb, and D. A. Scott, Performance Evaluation of Battery-Life-Aware Routing Schemes for Wireless Ad Hoc Networks, Proceedings of the IEEE International Conference on Communications, pp. 2824 2829, June 1997. [14] D. Kim, J. J. Garcia-Luna-Aceves, K. Obraczka, and P. M. J. Cano, Power-Aware Routing Based on the Energy Drain Rate for Mobile Ad Hoc Networks, Proceedings of International Conference on Computer Communications and Networks, pp. 565 569, Oct. 2002. [15] K. Tsudaka, M. Kawahara, A. Matsumoto, and H. Okada, Power Control Routing for Multi Hop Wireless Ad-hoc Network, Proceedings of IEEE GLOBECOM, pp. 2819 2824, Nov. 2001. [16] E. Jung and N. Vaidya, A Power Control MAC Protocol for Ad Hoc Networks, Proceedings of the Eighth Annual International Conference on Mobile Computing and Networking, pp. 36 47, Sept. 2002. [17] I. F. Akyildiz, J. Pelech, and B. Yener, A Virtual Topology Based Routing Protocol for Multihop Dynamic Wireless Networks, ACM-Kluwer Wireless Networks (WINET) Journal, vol. 7, no. 4, pp. 413 425, Aug. 2001. [18] The Network Simulator - ns-2. URL http://www.isi.edu/nsnam/ns/. 20