Multipath Contribution of Intermediate Nodes in AODV Extensions Abdulsalam Alammari, Ammar Zahary, Aladdin Ayesh University of Science and Technology, P.O.Box:13064, Sana’a, Yemen, De Montfort University, Leicester, LE1 9BH, UK. Emails: {a.alammari1, a.zahary}@ust.edu.ye,
[email protected]
Abstract—There are many multipath extensions to AODV (Ad Hoc on-demand Distance Vector) routing protocol in MANETs. These extensions can be categorized into two types based on the concept of utilizing multiple paths in intermediate nodes. The first type utilizes Multiple paths in Intermediate nodes in AODV protocol(MIAODV), and the other utilizes multiple paths only at source and destination nodes (in this paper, this type is called Non-MIAODV or NMIAODV). A comparison study is conducted in this paper for both types of multipath extensions to AODV. The traditional protocol AODV and the mechanism of Route Discovery Process (RDP) in both types MIAODV and NMIAODV is implemented using NS2. The performance is evaluated in terms of three performance metrics; routing packet overhead, packet delivery fraction, and average end-to-end delay. Keywords: AODV; MIAODV; NMIAODV; single path routing, multipath routing; intermediate nodes; route discovery.
I. I NTRODUCTION Multipath routing is a productive study domain in one of the most important challenge in Wireless Mobile Ad hoc Networks (MANETs) [1], multipath routing is appeared to resolve some single path problems that lead to some disadvantages, like routing packet overhead, end-to-end delay, data transmissions rates, power. Generally, multipath routing is considered as an advantage due to easy recovery from a route failure, and thus multipath protocols are considered more reliable and robust than single path protocols [2]. In a broad sense, multipath routing enables route reliability and also facilitates load balancing which are commonly used in several applications, especially in routing fault tolerance and Quality of Service (QoS) provisioning for heavy multimedia and real-time traffic. A single path routing means that a source node saves only one route (the optimal) in the routing table for a specific destination, if better route is detected, then the saved route will be changed to the better one. Ad hoc On-demand Distance Vector (AODV) [3], Destination Sequenced Distance Vector (DSDV) [4] and Wireless Routing Protocol (WRP) [5], are same examples of single path protocols. In multipath routing, all routes that are detected due to a route discovery are usually maintained in the source node routing table, thus Multipath protocols are considered more reliable and robust [2]. In multipath routing, the source node can select the optimal route among multiple available routes which enhances route availability and consequently minimizes frequent re-establishing of route discovery [6]. Dynamic Source Routing (DSR) [7], Temporally Ordered Routing Algorithm
978-1-4244-5844-8/09/$26.00 ©2009 IEEE
(TORA) [8] are the most common examples of traditional multipath routing protocols in MANETs. Both single path and multipath routing protocols in MANETs usually consist of two main processes (phases); Route Discovery Process (RDP) and Route Maintenance Process (RMP). This paper presents a comparison study for both types of multipath extensions to AODV. The contribution of intermediate nodes in multipath routing is experimented in multipath extensions to AODV for both types MIAODV and NMIAODV. The mechanism of MIAODV extensions is developed in this paper by implementing Route Discovery Process (RDP) in single path AODV which is achieved by receiving all Route REPlys (RREPs) and storing any numbers of routes to the same destination node. The destination node in turn can save multiple routes to the source node. In MIAODV extensions, the intermediate nodes can also save multiple reverse and multiple forward routes to both source and destination nodes respectively in RDP. In this paper, a performance comparison is presented for both types of AODV extensions when the intermediate nodes save multiple routes (As in MIAODV extensions), and when the intermediate nodes save only one reverse and one forward route (the optimal) to both source and destination nodes respectively (As in NMIAODV extensions). The mechanisms of MIAODV and NMIAODV extensions are implemented and the performances are evaluated against traditional AODV. The three protocols are implemented using NS2.30 and evaluated under the same simulation environment in terms of three performance metrics; average end-to-end delay, routing packet overhead, and packet delivery fraction. II. S INGLE PATH ROUTING IN AODV Based on the fact that AODV is a single path reactive protocol, nodes compute routes in AODV only when they are needed and only one route is selected (the optimal).Similar to all routing protocols in MANETs, AODV has two main processes RDP and RMP. This section explains in details the mechanisms of these two main processes in AODV routing protocol. A. RDP mechanism in AODV In AODV, the source node floods the RREQ packet in the network when a route is not available for the desired destination. It may obtain multiple routes to different
347
destinations from a single RREQ. The major difference between AODV and other on-demand routing protocols is that it uses a destination sequence number (DestSeqNum) to determine an up-to-date path to the destination. A node updates its path information only if the DestSeqNum of the current packet received is greater than the last DestSeqNum stored at the node. A RREQ carries the source identifier (SrcID), destination identifier (DestlD), source sequence number(SrcSeqNum), destination sequence number (DestSeqNum), broadcast identifier (BcastlD) and the time-to-live (TTL) field. DestSeqNum indicates the freshness of the route that is accepted by the source. When an intermediate node receives the RREQ, it either forwards it or prepares a RREP if it has a valid route to the destination. All intermediate nodes having valid routes to the destination, or the destination node itself, are allowed to send RREP packets to the source. AODV does not employ source routing of data packets. When a node receives a RREP packet, information about the previous node from which the packet was received is also stored in order to forward the data packet to the next node as it is the next hop toward the destination [1], [3]. III. R ELATED W ORKS OF M ULTIPATH AODV Single path abstraction is considered one of the most drawbacks of AODV that are extensively addressed in the recent studies of AODV extensions. Single path abstraction requires a source node to re-establish a new RDP when detecting a link failure in the primary current route. In traditional AODV, a RDP is invoked on-demand whenever the current route fails due to a link failure. A RMP starts by sending a RERR packet throughout the network. When a link failure is detected, a RERR is sent to all nodes related to that link to update their routing tables. In such case, a new RDP is invoked to obtain an alternative route. AODV selects only a single route (the optimal route) from all routes that are detected during a RDP. The mechanism of AODV increases frequent route rediscovery attempts and consequently increasing delay and control overhead. Many approaches are conducted to solve this problem of AODV either partial-route re-establishment or multipath establishment approaches. In partial-route re-establishment the routing protocol will find the alternative route in RMP, but in multipath establishment the routing protocol will save many routes during RDP phase. AODV Backup Routing (AODV-BR) [9] is an example of re-establishment approaches that try to find a partial-route as a backup when the routing protocol detects a broken link in the primary current route. Backup route is maintained at each neighbor of the primary current route to be used when needed which means not all intermediate nodes will save backup route. Only the nodes that are involved by a primary current route will save backup route. This also means only forward backup route can be offered while it does not take into consideration the reverse backup route. Finally, AODV-BR has a drawback of the limit number of routes. Multiple Next Hops (MNH) protocol [10] is another exten-
sion to AODV. It is an example of multiple-route establishment extensions that keeps track all nodes that send RREQ messages and wait for RREPs. At source side, multiple routes can be detected in case of receiving multiple RREP messages from multiple nodes. In any intermediate node, all duplicated RREQ are discarded (all new neighbors are registered toward the source node and RREQ re-broadcasting us discarded). In such case, multiple reverse routes can be offered, and in the case of forwarding multiple RREP, multiple forward routes can be offered. AOMDV [11] is another example of multiple-route establishment extensions to AODV. AOMDV detects multiple loopfree and link-disjoint routes. A notion of advertised-hop-count is used to guarantee loop-freedom of a route. In order to produce link-disjoint routes, a strong restriction is applied to route discovery process in AOMDV by the first-hop field in a RREQ packet, which is compared to the first-hop-list in a routing table. A vital drawback of AOMDV is that many efficient routes can be missed due to the restriction of link-disjoint routes, which leads to consume too much memory with increasing in routing overhead. One drawback with AOMDV is that it deletes links when they seem to be failed. The protocol sometimes considers congested links as broken links, and thus highly congested paths are removed by AOMDV mechanism. However, multiple forward and multiple reverse routes can be offered at any intermediate node. AODV Multipath [12] is another extension to AODV, which produces multiple routes with only node-disjoint feature. However, AODVM consumes too much memory with increasing in routing overhead because the node-disjoint restriction still produces inefficient routes. Because of AODVM is based on node-disjoint feature, only one reverse route and one forward route can be offered at any intermediate node. A recent extension to AODV is MRAODV [13] which is considered a non-disjoint multipath protocol. MRAODV reduces routing overhead by extending the waiting time of RREPs until detecting all possible routes included by all RREPs. In addition, MRAODV connects the separated reverse path fragments to help increasing number of multiple routes. However, MRAODV mechanism tends to wait for long time to check if there are more available routes including inefficient routes. In MRAODV multiple forward and multiple reverse routes can be offered at any intermediate node. Most recent multipath extensions to AODV are the Two Hops Backup Routing (2HBR) and AODV with Meshed Multipath (AODV-MM) [14]. The approach of 2HBR emphasizes on the RMP by detecting two hops alternative routes in case of a link failure. However, it leads to the original AODV performance if there is a failure in the backup route itself. In AODV-MM, 2HBR is extended and modified to improve packet delivery ratio by building meshed multiple alternative routes when receiving RREP packets. However, in 2HBR and its extension, AODV-MM, only multiple forward routes can be offered at each intermediate node. From the related work mentioned above, it is shown that the intermediate nodes can save multiple forward routes and mul-
348
tiple reverse routes only in multiple-route establishment extensions which are link-disjoint or non-disjoint types of route sharing. Three extensions MNH, AOMDV and MRAODV are examples of protocols that allow intermediate nodes to save multiple routes toward both source and destination nodes in RDP phase. The other extensions either offer forward backup routes such as AODV-BR, 2HBR and AODV-MM in RMP phase as they are re-establishment approaches. Also, they may offer only one reverse route and one forward route such as AODVM asthey are node-disjoint approaches. IV. MIAODV T YPE OF E XTENSIONS MIAODV type of extensions aims to optimize the routing overhead in RDP by benefit by flooding RREQ to detect multiple routes from the destination to the source, some intermediate node maybe will send RREP, and the destination will send RREP for every copy of RREQ reaches to it in case new route detected by this copy, then the source also will save multiple routes toward the destination, the intermediate nodes can also save multiple route toward each of source and destination nodes. This can be done using one RREQ and its corresponding RREP(s). MIAODV type aims to reduce routing packet overhead strive to enhance average end-to-end delay, and packet delivery fraction. In this paper, we present only the multipath routing mechanism of MIAODV type because the main difference between the mechanisms of MIAODV and NMIAODV types is that the intermediate nodes in NMIAODV type save only one reverse route and one forward route to each source and destination respectively in the network while they save multiple routs in MIAODV type.
B. RDP mechanism in MIAODV Type Route discovery process in MIAODV type is the same as in AODV with the following main differences: • The source node will send RREQ only if there is not a valid route to the destination. • When a node receives a duplicated RREQ (RREQ has the same broadcast id from the same source), the node will check if the previous hop for the RREQ packet is a new next hop toward the same RREQ source, thus it will add a new route to the RREQ source. This means that some nodes will have more than one route to the source node as shown in Figure 1 which shows RREQ broadcasting in MIAODV type. • If the node that received the RREQ isn’t the destination, it has to re-broadcast only one copy of RREQ message in case of no valid route is found to the destination. If a valid route is found, the node has to send only one RREP. If the node is the destination, it has to send a RREP for each new route and through the next hop of the new route. • When any intermediate node receives a RREP, it has to forward it through the route with the shortest hop count in case of finding many reverse routes. In case of finding only one route, a RREP will be forwarded through it. • When the source node receives many RREPs(generated by destination node or by any intermediate node), it will save many routes to the destination. Figure 2 shows RREP forwarding in MIAODV type.
A. Multipath routing mechanism of MIAODV Type MIAODV mechanism is structured to allow any node(source, destination, or intermediate nodes) to save multiple routes. MIAODV type (Based on RREQ flooding) allows any node (even the node is the destination or not) that hears duplicated RREQ to know about the new neighbor who re-broadcast this RREQ packet and saves a new reverse route toward the RREQ source. The destination node should send RREP for every new neighbor(which became the first hop in the new route) and through it. If any intermediate node has a valid enough route to the destination, it will send a RREP to the source node. This means the source node will receive many RREPs, thus it will save multiple routes to the destination. Some intermediate nodes may forward many RREPs which means that it can save multiple routes toward the destination node. Thus, the destination node will save multiple reverse routes to the source by receiving many copies of the RREQ packet. Also, some intermediate nodes will save multiple reverse routes to the source node. Because of the destination and the other nodes can send many RREPs, the intermediate nodes can save many forward routes to the destination node. Finally, the source will save multiple forward routes to the destination node. In order to add any route that can be detected at any node as new route, a new next hop should be found in the new route.
Fig. 1.
RDP in MIAODV routing protocol (request broadcast)
After adding a new route at any node, if there is more than one route toward the same destination, this node has to order these routes ascending based on hop count which means that the shortest one will be the primary route. V. S IMULATION E NVIRONMENT The simulation process is executed for traditional AODV and both types of extensions; MIAODV and NMIAODV under the same circumstances according to the most common performance metrics of routing evaluation in MANETs. In
349
•
•
•
Packet Delivery Fraction (PDF): the ratio of the data packets delivered to the destinations with respect to the sent packets. Average end-to-end Delay (AVGD) of data packets: delays caused by buffering, queuing, retransmission delay propagation time and transfer time. Normalized routing load (Routing Packets Overhead RPO): number of routing packets transmitted per data packet delivered at the destination. VI. R ESULTS S TUDY OF THE S IMULATIONS
Fig. 2.
In this section, firstly, the simulation results are presented for the evaluation of MIAODV and NMIAODV types against AODV protocol, and then the evaluation of MIAODV against NMIAODV type is presented. The simulation environment, input parameter, and performance metrics used in this section are described in the previous section.
RDP in MIAODV routing protocol (replay forwarding)
this simulation, MIAODV type represents the mechanism of all multipath extensions to AODV that utilize multiple routes in intermediate nodes where NMIAODV type represents the mechanism of all extensions that utilize only one route at intermediate nodes and multiple routes at source and destination nodes. To execute the simulations, the same mobility models and traffic scenarios have been used in NS2 environment for all simulations. In this section, the mobility, traffic scenarios, and performance metrics used in the simulation process are described firstly and then, the results study and the performance evaluation are presented later in the next section.
A. Results study of MIAODV and NMIAODV against AODV The results study of AODV protocol and both types MIAODV and NMIAODV is presented here showing the differences between AODV and both types of extensions. The simulation is executed by simulating 30 connections between 100 nodes as an example of the difference of performance in the traditional AODV and any type of its extensions. Figure 3 illustrates AODV performance versus mobility
A. Mobility models and traffic scenarios The mobility models used are based on the random waypoint model in a rectangular area. Coverage area used for the simulation is 500m x 500m with 100 nodes scattered randomly. The maximum speed of the nodes is set to 20m/s, node pauses for a while and then removes to another random location within the specified area. Simulations are run for 250 seconds and the pause time have been varied as 0s, 10s, 20s, 30s, 40s ,50s ,100s and 250s. Protocols are simulated using different mobility and traffic scenarios for more accurate results. Constant Bit Rate (CBR) sources each of 512B packet size are used for traffic and the wireless standard IEEE 802.11b with 11Mbps transmission rate is used for MAC sub-layer. In addition to traffic scenarios, mobility is one of the most important input parameter in this paper which affects the behavior of routing protocols. Pause time is used to measure the degree of mobility. Small pause time value indicates to a high mobility thus, if a pause time = 0s, this means that nodes continue moving and do not stop. Large pause time value indicates to a low mobility thus, if a pause time = 250s, this means that nodes stop for long time (250s) before start moving again. B. Performance metrics The performance metrics used to evaluate our simulation results can be summarized as follows:
Fig. 3.
Average PDF of AODV against MIAODV and NMIAODV
compared to MIAODV and NMIAODV types in terms of packet delivery fraction (PDF) at 100 nodes 30 conn. Simulation results show that MIAODV and NMIAODV types are clearly better than AODV in all mobility scenarios. By comparing MIAODV and NMIAODV types, the simulation results show that MIAODV type is better than NMIAODV type in all mobility scenarios except at PT = 10 and 250 seconds where NMIAODV type became slightly better. Generally, Figure 3 shows clearly that AODV has the lowest performance in all mobility scenarios in terms of PDF. Figure 4 illustrates AODV performance versus mobility compared to MIAODV and NMIAODV types in terms of average end-to-end delay (AVGD) at 100 nodes 30 conn. Simulation results show that MIAODV and NMIAODV types are clearly better than AODV in all mobility scenarios. By comparing MIAODV and NMIAODV types, the simulation results show that MIAODV type is better than NMIAODV type in all mobility scenarios except at PT = 30 and 250
350
Fig. 4.
Average AVGD of AODV against MIAODV and NMIAODV
seconds where NMIAODV type became slightly better. Generally, MIAODV is the best in terms of AVGD while NMIAODV performs better than AODV in all mobility scenarios.
Fig. 5.
Average RPO of AODV against MIAODV and NMIAODV
Figure 5 illustrates AODV performance versus mobility compared to MIAODV and NMIAODV in terms of routing packets overhead (RPO) at 100 nodes 30 conn. Simulation results show that MIAODV and NMIAODV types are clearly better than AODV in all mobility scenarios. By comparing MIAODV and NMIAODV types, the simulation results show that MIAODV type is better than NMIAODV type in all mobility scenarios except at PT = 0, 50, and 250 seconds where NMIAODV type became slightly better. Figure 5 shows clearly that AODV has the lowest performance in all mobility scenarios in terms of RPO.
that more scenarios are considered with varied number of connections (20, 30, 50 and 70). In addition, the scenarios are divided into three groups based on the pause time in order to study the results easily. These groups are as follows: • High mobility which is indicated by pause time = 0. • Medium mobility which is indicated by pause time = 10, 20, 30, and 50. • Low mobility which is indicated by pause time = 100 and 250. Each group is studied for each number of connections described above, thus each group value in the figures represents the average performance at all pause times involved by the corresponding group at a specific number of connections. Bellow is the performance study of MIAODV against NMIAODV type in high mobility scenario (pause time = 0) for the set of number of connections described above. Figure 6 illustrates the performance of MIAODV compared to NMIAODV versus mobility in terms of Packet Delivery Fraction (PDF). Simulation results show that in high mobility, MIAODV outperforms NMIAODV at medium number of connections (30 and 50) where NMIAODV became better in the case of small and large number of connections (20 and 70). As shown by the simulation results of medium mobility, MIAODV outperforms NMIAODV at number of connections (20 and 30) where NMIAODV became better at the other numbers of connections. In low mobility, it is shown that MIAODV outperforms NMIAODV at medium number of connections (20 and 30) while NMIAODV became better in the other numbers of connections except at the number 70 where MIAODV still better in terms of PDF.
Fig. 6.
Average PDF of MIAODV against NMIAODV in all scenarios
B. Results study of MIAODV against NMIAODV The results study in the previous subsection presents the enhancement that added to AODV in both types of its multipath extensions which clearly shown by comparing AODV results to the results of both MIAODV and NMIAODV types in all mobility scenarios in terms of the three performance metrics. However, the results study mentioned above does not clearly answer the important question in this paper. The question is ”Which is better for intermediate nodes, to save multiple routes or a single route?”. To answer that question, more simulations are executed with more scenarios for both MIAODV and NMIAODV types using the same simulation environment, input parameter, and performance metrics used in the simulations presented above. The difference here is
Figure 7 illustrates the performance of MIAODV compared to NMIAODV versus mobility in terms of Average end-to-end Delay (AVGD). Simulation results show that in high mobility, MIAODV outperforms NMIAODV at medium number of connections (30 and 50) where NMIAODV became better in the case of small and large number of connections (20 and 70). In medium mobility, simulation results show that MIAODV outperforms NMIAODV at all number of connections except at number 50 where NMIAODV became better. As shown by the simulation results of medium mobility, NMIAODV performs better than MIAODV in medium number of connections (20 and 30) while MIAODV became better in the other numbers of connections.
351
Fig. 7.
Average AVGD of MIAODV against NMIAODV in all scenarios
Figure 8 shows the performance of MIAODV compared to NMIAODV versus mobility in terms of Routing Packets Overhead (RPO). Unlike PDF and AVGD, MIAODV type performs better for high mobility scenarios in the case of small and large number of connections (20 and 70) while NMIAODV became better at medium number of connections (30 and 50). In high mobility, it is generally shown that the performance of MIAODV compared to NMIAODV has a reverse behavior in terms of RPO compared to PDF and AVGD at a particular number of connections. The higher performance of MIAODV and NMIAODV type in terms of RPO, the lower performance in terms of both PDF and AVGD and vice versa. In medium mobility, the simulation results show that MIAODV performs better than NMIAODV at number of connections (20 and 30) while NMIAODV became better in the other numbers of connections. Generally, the lower RPO performance in medium mobility leads to positive effect for MIAODV and NMIAODV in terms of PDF and AVGD. As shown by the simulation results of medium mobility, MIAODV outperforms NMIAODV in medium number of connections (30 and 50) where NMIAODV has better or equal performance in the other numbers of connections. Unlike high and medium mobility where RPO has a reverse effect against PDF and AVGD, it is clear generally in low mobility that AVGD has a reverse effect against PDF and RPO. The lower AVGD performance in low mobility leads to positive effect for MIAODV and NMIAODV in terms of PDF and RPO. Therefore, we can conclude here the inverse effect of RPO against both PDF and AVGD in high and medium mobility scenarios at all numbers of connections simulated above. On the other hand, we can conclude the inverse effect of AVGD against both PDF and RPO in low mobility scenarios at all numbers of connections simulated above.
Fig. 8.
Average RPO of INAODV against NMIAODV in all scenarios
VII. C ONCLUSION This paper presents a comparison study for two types of multipath extensions to AODV; multipath intermediate (MIAODV) and non-multipath intermediate (NMIAODV) extensions. The contribution of intermediate nodes in multipath routing is experimented in multipath extensions to AODV for both types and the mechanism of each type is developed in this paper by implementing route discovery process in traditional AODV and both types of AODV extensions; MIAODV and NMIAODV. Simulation results show that both types of AODV extensions extremely outperform traditional AODV. On the other hand, MIAODV type outperforms NMIAODV type in most mobility scenarios, especially in high and medium mobility. The evaluation is performed for the performance of each type in terms of three performance metrics; packet delivery fraction, average end-to-end delay, and routing packets overhead. As a future work of this research, a simulation process can be executed and tested for the implementation of route maintenance process of each type. In addition, a testing process can be performed in a wide classification of AODV extensions such as node-disjoint, link-disjoint, and non-disjoint extensions along with the features of MIAODV and NMIAODV. R EFERENCES [1] Ilyas, M. The Handbook of Ad Hoc Wireless Networks, Florida Atlantic University, Boca Raton, Florida, CRC Press, 2003. [2] Parissidis, G., Lenders, L., May, M., Plattner, B. Multipath Routing Protocols in Wireless Mobile Ad Hoc Networks: A Quantitative Comparison,NEW2AN 2006, 2006. [3] Perkins, C., Belding-Royer, E., and Das, S., D.A. Ad-hoc on-demand distance vector (AODV) routing, 2003. [4] Perkins, C.E., Watson, T.J. Highly dynamic destination sequenced distance vector routing (DSDV) for mobile computers, ACM SIGCOMM’94 Conference on Communications Architectures, London, UK, 1994. [5] Murthy, S. J.J. and Garcia, L. A. A routing protocol for packet radio networks, Proceedings of the First Annual ACM International Conference on Mobile Computing and Networking,Berkeley, CA, 1995. [6] Liu, M., Xu, Z., Yang, J., Ye, J. Collision-constrained Minimum Energy Node-disjoint Multipath Routing in Ad Hoc Networks, Proceedings of IEEE, Wuhan City, China, 2006. [7] Johnson, D.B. and Maltz, D.A. Dynamic Source Routing in Ad Hoc Wireless Networks, Mobile Computing, Kluwer Academic Publishers, The Netherlands, 1996. [8] Park, V. D. and Carson, M. S. A highly Adaptive Distributed Routing Algorithm for Mobile Wireless Networks, Proc. INFOCOM 97, Kobe, Japan, 1997. [9] Lee, S. J. and Gerla, M. AODV-BR: Backup Routing in Ad hoc Networks, Proc. of IEEE Wireless Communications and Networking Conference, Chicago, IL, USA, 2000. [10] Jiang, M.H. and Jan, R.H. An Efficient Multiple Paths Routing Protocol for Ad-hoc Networks, Proc. of the 15th International Conference on Information Networking, 2001. [11] Marina, M. K. and Das, S. R. On-demand Multipath Distance Vector Routing in Ad Hoc Networks, Proceedings of the 9th IEEE International Conference on Network Protocols (ICNP), Mission Inn, Riverside CA, USA, 2001. [12] Ye, Z., Krishnamurthy, S. V., Tripathi, S.K. A Framework for Reliable Routing in Mobile Ad Hoc Networks, IEEE INFOCOM, 2003. [13] Higaki, H. and Umeshima, S. Multiple-Route Ad hoc On-Demand Distance Vector (MRAODV) Routing Protocol, The 18th International Parallel and Distributed Processing Symposium (IPDPS04), IEEE, Santa Fe, New Mexico, USA, 2004. [14] Kuo, C. T., Liang, C. K. A Multipath Routing Protocol in Mobile Networks, International Conference on Information Technology: Research and Education, IEEE, 2006
352
