INTERNATIONAL JOURNAL OF APPLIED ENGINEERING RESEARCH Volume 1, No1, Copyright 2010 All rights reserved Integrated Publishing Association

Transcription

1 Multipath Dynamic Source Routing with Cost and Ant Colony Optimization for MANETS Sarala.P 1, Kalaiselvi.D 2 1. Lecturer 2. Post graduate student in Computer Science and Engineering Nandha Engineering College, Erode, Tamilnadu saralagokul@yahoo.com ABSTRACT The mobile adhoc networks are the infrastructureless networks constructed without any fixed infrastructure such as base station, tower, redirection switches and routers. Mobile adhoc networks are the temporary wireless networks. All the routing information are managed by the node itself. Mobile adhoc network routing operations are categorized into two types proactive and reactive routing. The multipath routing mechanisms are used to discover multiple paths under the nodes. Multipath dynamic source routing protocol (MP DSR) is used to discover multipath route under MANET nodes. The MP DSR protocol uses the local link information for the route discovery process. The MP DSR protocol is enhanced with ant colony optimization method to provide multipath route information using global link information. EMP DSR provides QoS factors such as end to end reliability. Network traffic, bandwidth and battery power factors make an influence over the route discovery process. Cost enabled route discovery is one of the consierable routing method that enables the cost estimation with different metrics. The multipath routing protocols concentrates on the route discovery with end to end reliability factors. The EMP DSR protocol is integrated with fuzzy cost estimiation techniques. Distance, network traffic, bandwidth and battery power metrics are used in the fuzzy cost enabled multipath dynamic source routing protocol. Keywords: Fuzzy network, Ant colony algorithm, adhoc networks, multipath routing. 1. Introduction Mobile Ad hoc Networks (MANETs) consisting of a collection of wireless mobile hosts (called nodes) have received increasing attention recently. Independence from central network administration, ability to be self configured, self healing through continuous reconfiguration, scalability and flexibility are the distinguished reasons to deploy such networks [1]. Routing as one of the cornerstones of any network including MANETs, is needed whenever data packets need to be handed over several nodes to arrive at their destinations. Routing protocols have to find routes for packet delivery and make sure the packets are delivered to the correct destinations [7]. In MANETs each node serves as a router to forward packets to other hosts. Since, nodes are free to move randomly and organize themselves arbitrarily; the network s topology may change rapidly and unpredictably. Such characteristics allow an ad hoc network to be established on the fly with built in fault tolerance and unconstrained connectivity. This makes routing in such networks more challenging, especially when certain Quality of Service (QoS) requirements are to be guaranteed during the routing. One of the most important QoS metrics is to provide end to end reliability. End to end reliability is used to reflect the 83

2 probability of sending data successfully from the source node to the destination node within a time window. End to end path reliability is the product of the link availabilities along the path, assuming the link availabilities are independent. The calculation of link availability is based on the current s node movement. The words availability and reliability will be used interchangeably throughout the paper. Usually required end to end reliability for routing is satisfied through applying multi path scheme. Taking this scheme into consideration, the endto end reliability is 1 Π k K (1 k), where k is the reliability of a path, and K is the set of all paths to be considered in the reliability calculation. End to end reliability guarantees that there is an active path between two mobile nodes within a time interval. In this paper, our major contribution goes towards supporting the reliability as a QoS metric through multi path routing. Many researches have been done to improve QoS in multipath routing algorithms for MANETs [2][10][11]. However, only few of them have focused on reliability metric. The most related work to ours is Multi Path Dynamic Source Routing (MP DSR), which is based on Dynamic Source Routing (DSR) protocol. MP DSR seeks to compute a set of unicast routes that can satisfy a minimum end toend reliability requirement. It then maintains this requirement throughout the lifetime of transmission. However, the route discovery in MP DSR relies only on local link availability information at each intermediate node to perform the route request message forwarding, without using any global information. Selecting a reliable link in an intermediate node solely based on local information may not necessarily lead to finding a satisfactory reliable end toend path. The problem originates from the fact that there is no global end to end reliability related information available for each node. To alleviate the aforementioned problem and provide the global reliability related information for each node, in this paper an Enhanced Multi Path Dynamic Source Routing algorithm called EMP DSR has been developed. To this end, we have applied an Ant Colony Optimization (ACO) based approach called AntNet, a lightweight optimization method. Generally ACO algorithms utilize ants, which are mobile agents that are capable of solving various optimization problems such as network routing and congestion problems [5]. 2. Preliminaries The fundamental concepts and approaches which our proposed method is based upon are introduced to the necessary extent. First MP DSR method then a brief introduction of Ant Colony Optimization and AntNet are given MP DSR MP DSR or Multi Path Dynamic Source Routing introduced, tries to find multiple disjoint paths from a given source to a destination while guaranteeing that these paths altogether can satisfy a given end to end reliability P u where 0 < P u 1. MP DSR achieves route discovery by determining the number of paths to be discovered (m 0 ) and the lowest reliability (Π lower ) that each of the m 0 paths must have so that the P u reliability can be satisfied. Once, obtaining the appropriate values for the parameters at the source node, the source node sends m0 Route Request (RREQ) messages to search for feasible paths. Each of the RREQ messages contains useful information such as, the path it has traversed till then (pathvector), 84

3 pathvector s corresponding accumulated reliability namely Π acc and Π lower. When an intermediate node receives an RREQ message, it investigates whether Π acc Π lower or not. If this inequality fails, the message is discarded. Otherwise, it appends itself to pathvector and calculates the new Π acc and finally forwards at most m 0 copies of the modified message to its neighbors. After a while, the destination node collects some of the RREQ messages. From the pathvectors stored inside RREQ messages, the destination node uses a path selection algorithm to pick the set of disjoint paths that can satisfy P u. Two paths are disjoint if they just share common source and destination nodes, but do not share common intermediate nodes. Then, the destination node responds to the source by sending Route Reply (RREP) messages through the selected paths. Thereafter, the source node commences sending the packets through these paths Ant Colony Optimization (ACO) ACO, a famous swarm intelligence approach, has taken its inspiration from the social behaviors of real world ants. Most often real ants are wandering stochastically around their nests to forage (Search for food). Upon finding food, they return back to their nests and simultaneously deposit pheromone trails along the paths. Since ants tend to follow the pheromone trails, they more likely biased towards such paths, and as a result they may not keep on traveling quite randomly. Therefore, they likely move through these paths and reinforce the existent pheromone. This kind of indirect communication is called stigmergy [4] in the biology and entomology literatures. Over time, however, the pheromone trail starts to evaporate, thus reducing its attractive strength. Analogously, ACO [4], one of the state ofthe art paradigms in designing meta heuristic algorithms for combinatorial optimization problems, utilizes artificial mobile agents namely ants which are capable of solving various kinds of routing and congestion problems. At regular intervals from every network node several ants are launched toward the destination node to discover the feasible low cost path to that node. Each ant in ACO considers two parameters to select its next hop. The first one is the amount of pheromone deposited on the path to the next node, and the other is a kind of heuristic parameter such as the queue length associated with the link. AntNet as an ACO approach to adaptive learning of routing tables in communication networks. Each node k in the network stores some data structures within itself which are responsible for keeping local traffic statistics, and routing table. Local traffic statistics defines a simple parametric statistical model for traffic distribution over the network as seen by node k. In fact, it explains the amount of traffic flows towards each possible destination. Routing table, for each possible destination d and for each node n, stores a probability value P nd which expresses the desirability of selecting n as the next node when the destination node is d. In fact it shows amount of pheromone deposited on the link (k,n). When an ant at node k heads toward a destination node d, it selects the next neighbor node n with the probability P nd where we have: P nd = P nd + α l n / 1 + α ( N k 1) (1) Here l n = 1 q n / Σ Nk n =1 qn (2) Where N k is the number of the neighbors of node k, q n is the length of the queue associated with the link connecting k to n and α is the weight of the importance of the heuristic function with respect to pheromone deposit. When an ant reaches the destination node, this node can now evaluate the goodness of the path. This goodness can be de fined according to an application s requirement [8]. AntNet itself uses the ants trip time and the 85

4 parameters of the local statistical model. Quantifying the goodness value, the destination node creates a backward ant. The backward ant takes the exact path by the corresponding ant, but in the opposite direction, and deposits an amount of pheromone on its path to the source node. This amount is commensurate with the goodness of this path. 3 Related Work On demand routing protocols generally perform well for wireless ad hoc networks, since the flooding of route request messages is only performed when a route is needed, rather than periodically as in proactive routing protocols. The degree of flooding is further reduced by using multi path routing protocols, which have been proposed to discover multiple paths for data transmission. Such protocols can be considered as a hybrid of proactive and ondemand routing, because route discovery is invoked on demand while route maintenance is done on a proactive basis. Examples of such multi path protocols include Temporally Ordered Routing Algorithm (TORA) [9] and Split Multi path Routing (SMR) [6]. In TORA, the source node constructs multiple routes by flooding a query message followed by a set of update messages. However, TORA does not have any mechanisms to evaluate the quality of these multiple paths and this leads to its poor performance. MP DSR overcomes this problem by selectively choosing more reliable paths and by providing soft guarantees on the end toend reliability. SMR [6] extends DSR in the way that the destination can discover two paths for each route request, in which one is the shortest path, and the other is the maximum disjoint path. There is no explicit enforcement of disjoint paths and this differs from our work, because our algorithm enforces the use of disjoint paths in its route discovery in order to use the definition of path reliability to provide end to end reliable service. Previous work in QoS routing for ad hoc wireless networks focuses on guarantees with respect to bandwidth, cost and delay. One of such routing protocols is the ticket based QoS routing protocol [3]. It considers two kinds of routing criteria: the delay constrained least cost routing and the bandwidth constraint leastcost routing. It uses ticket based probing to control the number of route queries and to find multi path in parallel. In comparison, our MP DSR considers the dynamic nature of network topology as well as the importance to offer continuous network connection in certain mission critical applications. Thus, the objective of our protocol is to improve the level of service by providing guarantee with respect to end to end reliability, and to probabilistically guarantee the required connection lifetime. In addition, our MP DSR differs in the way of searching multiple paths; the route discovery in our protocol relies only on local link availability information at each intermediate node to perform the route request (RREQ) message forwarding, without resorting to any global information as was used in [3]. 4. EMP DSR 4.1. Motivation MP DSR tries to compute a set of unicast routes that can satisfy a minimum end toend reliability requirement. It then maintains this requirement throughout the lifetime of transmission. MP DSR is fed with just local information for discovery of routes. Following a high reliable link, which connects the current node to one of its neighbors, cannot guarantee an end to end reliable path, since the selected neighbor may not have good reliable links towards destination node. As a result the obtained path may not have the end to end 86

5 satisfactory reliability. To mitigate this problem, we have used AntNet running silently in background to provide the required global information for MP DSR Ants' Contribution to EMP DSR EMP DSR uses a modified version of AntNet to populate its own routing table. The most significant change made to AntNet goes towards the way the goodness value is obtained. In our approach, we have used end to end reliability as the goodness parameter. A forward ant starts the trip to the destination node, then it arrives at an intermediate node, where it needs to update its Π acc field, by multiplying Π acc with the link availability of the link the forward ant has just traversed. This process continues until the ant eventually reaches its destination node. As a result, Π acc at the destination is the end to end traversed path reliability. Using Π acc and trip time (T) obtained from the forward ant; the paths goodness is calculated from equation 3: PathGoodness = c 1 (Π acc ) + c 2 (W best /T) (3) In equation (3), W best is the best trip time experienced by the forward ants traveling toward the destination over the last observation window. The maximum size of the window is set to a constant, for the sake of simplicity. The coefficient c 1 and c 2 weight the importance of each term. In the current implementation of the algorithm c 1 = 0:6 and c 2 = 04. The first term is the end to end path reliability. The second term implies the goodness of the round trip time (T) relative to recent best trip time (W best ). The first term is the most important one, while the second term alleviates the stagnation problem [4], by conducting some of the ants to less reliable but less congested or shorter paths. Stagnation occurs when ants are attracted by one optimal path which will lead this path to be heavily congested. After the forward ant reaches the destination node, the destination node generates a backward ant, transfers the entire forward ant s memory to it. The backward ant takes the same path as that of its corresponding forward ant, but in the opposite direction. The backward ant updates the routing table at intermediate nodes for all the entries related to the forward ant s destination node. In this update the related probabilities will be increased by a value proportional to both path goodness and the previous value of the probability (P nd ) through equation 4. P nd = P nd + PathGoodness (1 P nd ) (4) Launching several ants at regular intervals at different nodes, the whole process silently runs in the background all the time and updates routing tables Path Discovery Upon receiving a connection request from an application with a certain reliability criterion (P u ), the source node initiates the PathDiscovery algorithm with the given P u (illustrated in Algorithm 1). PathDiscovery process needs to determine m 0 (number of paths), Πlower (the minimum reliability that each of the m 0 paths requires to guarantee P u ), and t w (the time window that this end to end reliability holds within). Π lower is calculated from the given P u and m 0 through equation 5: Π lower = 1 m 0 1 P u (5) In PathDiscovery, tw is set to the constant t wmax (here 100 seconds), the upper bound of the time window. m 0 is initiated to 1 from which Π lower is calculated (line 8). Then, procedure iterates through all links associated with the neighbors of the source node and 87

6 checks if there are m 0 number of links with at least Π lower reliability (lines 9 13). Basically, the procedure tries to find the minimum number of links which altogether can satisfy P u. There is also a maximum threshold (m max ) for number of disjoint paths to be discovered (line 5). Thus, if m0 violates m max or the number of neighbors (L), the process will need to decrease t w due to the fact that links can be more reliable in the shorter time window (line 31). However, it is preferred not to have smaller time windows, since source node is required to send route check messages to validate whether the end to end reliability can still satisfy P u or not. This validation incurs extra overhead especially when it is done at short intervals periodically. Therefore, a minimum threshold is adopted for t w called t wmin (here 20 seconds), if the algorithm fails for the minimum threshold then it is a failure for the application request (line 30). Otherwise, the source node will send RREQ messages to the m 0 highest reliable links. The highest reliable links are the ones who have the highest value in ant generated table for the requested end destination (lines 14 28). Note that MP DSR s link selection relies on local information, whereas EMP DSR utilizes the more global and consequently worthwhile information about the network status provided by ants. Upon receiving an RREQ message, each intermediate node runs HandleRREQMessage Algorithm (illustrated in Algorithm 2). In this algorithm, the intermediate node first checks whether it is the destination itself or not. If not then the HandleRREQMessage algorithm will try to forward the RREQ towards the destination. An intermediate node only forwards m 0 number of the same RREQ messages, and the same RREQs will be discarded. This is accomplished via the bookkeeping variable numofforwardedmsg (lines 1 2). In order to keep the required reliability criterion along the path, the RREQ contains a reliability field Π acc that keeps reliability value from source to the current node. Each time it passes a link, it multiplies Π acc with the links associated reliability it has just traversed (line 4). The procedure then calculates L i (t w ) according to the equation 6 (line 5). L i (t w ) = Π lower / Π acc (6) With the current Π acc, L i (t w ) is the minimum reliability that the rest of path must hold in order to keep the end to end path reliability greater than or equal to P u. Then, the highest reliable links retrieved from ant populated table, which also satisfy the L i (t w ), are picked and the RREQ messages are forwarded to them (lines 8 17). To avoid loops, the selected neighbors must not be visited before (lines 9 10). Algorithm 1 Path Discovery (P u ) 1: t w = t w Max; 2: while t w t wmin do 3: set A A s,1(t w ),..., A s,t(t w ); 4: m 0 = 0; 5: while m 0 L and m 0 m max do 6: path = 0; 7: m 0 = m 0 + 1; 8: Π lower = 1 m 0 1 P u ; 9: for neighborj seta do 10: if A s,j (t w ) Π lower then 11: path = path + 1; 12: end if 88

7 13: end for 14: if path m 0 then 15: Π end_to_end = {(n 1, v 1 ),.., (n n,v n )}; {where the list is sorted according to v i } 16: rreq = newrreq(m 0,Π lower,t w ); 17: n = numberofneighbors of source node; 18: numrreqmsg = 0; 19: for i = 1 to i = n do 20: if A s (t w ) Π lower then 21: Send(RREQ;Π endtoend[i].neighbor); 22: numrreqmsg + +; 23: if numrreqmsg = m 0 then 24: return success; 25: end if 26: end if 27: end for 28: end if 29: end while 30: return error 31: t w = 0.9 t w ; 32: end while When the destination node receives the first RREQ, it triggers a timer to somehow limit the time that should wait for the same consecutive RREQs. The more the time limit, the greater number of RREQs are received at the destination. On the other hand, higher time limits increase path discovery delay. Thus, it is better to set the time limit according to network size. As soon as the timer time outs, the destination node launches its path selection algorithm. Each of the RREQ messages contains a path with at least Π lower reliability. Then, the destination node selects a set of these paths, so that these selected paths (TraceSet) can collectively satisfy P u. To attain this, it sorts all the available paths regarding to their reliability in descending order, and names this sorted paths CandidateSet. Path selection algorithm intrinsically is performed recursively. At each step, the algorithm picks a new path from CandidateSet and tries to add it to the TraceSet upon condition that the new path is disjoint with respect to the previously added paths in TraceSet. Then, the algorithm recursively invokes itself and attempts to add further paths. If none of the paths remaining in the CandidateSet were eligible to be included in TraceSet, the algorithm removes the previously added paths from TraceSet, and backtracks to look for other possible paths from CandidateSet. The whole procedure continues until either the algorithm finds an adequate number of paths to satisfy desired reliability, or exits with no achievement. Algorithm to handle RREQ Message ( RREQ) 1: if numofforwardedmsg[rreq] > m 0 then 2: return ; 3: end if 4: Π acc = Π acc A j,i (t w ); 5: L i (t w ) = Π lower / Π acc ; 89

8 6: n = num of the neighbors of the intermediate node; 7: P end_to_end = {(n 1, v 1 ),, (n n, v n )}; {where the list is sorted according to vi} 8: for k = 1 to n do 9: if P end_to_end [k].neighbor pathvector then 10: continue; 11: end if 12: if A i,k (t w ) L k (t w ) then 13: numofforwardedmsg[rreq] + +; 14: forward(rreq, P end to end [k].neighbor); 15: return ; 16: end if 17: end for 5. Fuzzy Cost Estimation on EMP DSR The proposed system improves the EMP DSR protocol with cost based route discovery scheme. The fuzzy logic techniques are used for the cost estimation. Bandwidth and traffic factors are used in the route discovery process. The proposed system improves the end to end reliability. Multi path route discovery is designed in the system. Cost factor is used for route discovery process. Dynamic source routing protocol is used in the system. Alternate route is selected in node failure state. The system is divided into three major modules. They are Route discovery, Route maintenance and Cost estimation. 5.1 Route Discovery The route discovery module is developed to find out path between given source and destination nodes. The route request is passed to all nodes. The neighbor list updated using route reply. The system produces multiple paths for the source to destination node. 5.2 Route Maintenance The route maintenance module is designed to perform route update process. The node failure and node mobility are monitored in the route maintenance process. The path is updated with respect to current network status. The system maintains active path for data transmission. 5.3 Cost Estimation The system uses the cost factor for route selection process. The fuzzy logic technique is used for cost estimation process. Bandwidth, traffic, node count and battery power factors are used in cost estimation. The system improves the reliability using multiple paths. 90

9 6. Conclusions Multipath routing supports end to end reliability. The Enhanced Multi Path Dynamic Source Routing (EMP DSR) is improved with fuzzy cost mechanism. The bandwidth and traffic factors are included in the route discovery process. The system provides user choice based route discovery process. Cost based route estimation. High reliability. Multipath routing. Global optimized route discovery. 7. References 1. Breed.G, "Wireless ad hoc networks: Basic concepts. In High Frequency Electronics", pages 44 46, Chen. Y.S, Tseng.Y.C, Sheu.J.P, and Kuo.P.H. "An ondemand, link state, multi path QoS routing in a wireless mobile ad hoc network". In European Wireless, Chen.S and Nahrstedt. K "Distributed Quality of Service Routing in Ad Hoc Networks". IEEE Journal on Selected Areas in Communications, 17(8), August Dorigo.M and Stutzle.T "Ant Colony Optimization". MIT Press, Ehsan Khosrowshahi Asl, Morteza Damanafshan, Maghsoud Abbaspour Majid Noorhosseini, Kamran Shekoufandeh "EMP DSR: An Enhanced Multi Path Dynamic Source Routing Algorithm for MANETs Based on Ant Colony Optimization" 2009 Third Asia International Conference on Modelling & Simulation IEEE. 6. Lee. S and Gerla.M "Split Multipath Routing with Maximally Disjoint paths in Ad Hoc Networks". In Technical Report in University of California, Mueller.S, Tsang.R.P, and Ghosal.D. "Multipath routing in mobile ad hoc networks: Issues and challenges". In MASCOTS Tutorials, pages , OMNeT++. January Park.V and Corson. S "Temporaly ordered routing algorithm". Internet Draft, August Tsirigos.A and Haas.Z "Multipath routing in the presence of frequent topological changes". IEEE Communications Magazine, 39(11), Valera.A.C, Seah.W.K.G, and Rao.S.V. "Cooperative packet caching and shortest multipath routing in mobile ad hoc networks". In INFOCOM. IEEE Computer Society,