PaperZone

Preface An ad hoc network is a collection of mobile nodes that dynamically form a temporary network. Unlike traditional network, that the nodes use the infrastructure to form the network, an ad hoc network operates without the support of any infrastructure. Zone routing protocol [1] is a widely used protocol for ad hoc network. It is the most promising and widely accepted protocol in mobile ad hoc network (MANET). Recently there has been some study based on Genetic algorithm (GA) [2] that uses the zone routing protocol as an application and performs better than the traditional method [3][4][5]. Estimation of distribution algorithm [6] is an evolutionary method based on probabilistic model, for searching and optimization. {/** this line comes from nowhere and ends with no hints: there is no relationship mentioned between this EDA and zone routing protocol. Something like: as EDA is … …, it might …. in zone routing protocol.**/} The objective of our thesis will be the use of zone routing protocol as an application in estimation of distribution (EDA) algorithm (EDA), and compare the performance with the performances of the traditional method as well as the method used by the genetic algorithm. The rest of the proposal is organized as follows: chapter 1 describes the genesis of zone routing protocol as an application in genetic algorithm and the objective of our thesis. Chapter 2 describes briefly some of the background topics briefly needed for the purpose of the thesis. Chapter 3 describes the motivation and research focus of our thesis as well as and the research methodology and the our proposed method with the out line of implementation detail of the our thesis. Chapter 4 will give a summery summary of the study and the idea about the future work. {/** insert a page break here: it will ensure the format of the following sections **/} Chapter 1 INTRODUCTION 1.1 Genesis In the next generation of wireless communication systems, there will be a need for the rapid deployment of independent mobile users. Significant examples include establishing survivable, efficient, dynamic communication for emergency/rescue operations, disaster relief efforts, and military networks. Such network scenarios cannot rely on centralized and organized connectivity, and can be conceived as applications of Mobile Ad Hoc Networks MANETs. A MANET is an autonomous collection of mobile users that communicate over relatively bandwidth constrained wireless links. Since the nodes are mobile, the network topology may change rapidly and unpredictably over time. The network is decentralized, where all network activity including discovering the topology and delivering messages must be executed by the nodes, i.e., routing functionality will be incorporated into mobile nodes. Mobile ad hoc networks MANETs became a popular subject for research as laptops and 802.11/Wi-Fi wireless networking had become became widespread in the mid to late 1990s. Many of the academic papers evaluate protocols and abilities assuming varying degrees of mobility within a bounded space, usually with all nodes within a few hops of each other, and usually with nodes sending data at a constant rate. Different protocols are then evaluated based on the packet drop rate, the overhead introduced by the routing protocol, and other measures. 1.1. 1 Zone Routing Protocol Among all the protocols used in MANETs, Zone Routing Protocol (ZRP) [1] is the most popular and widely accepted routing protocol. Zone Routing Protocol or ZRP was the first hybrid routing protocol with both a proactive and a reactive routing component. ZRP was first introduced by Haas in 1997 [7]. ZRP is proposed to reduce the control overhead of proactive routing protocols and to decrease the latency caused by routing discover in reactive routing protocols. ZRP defines a zone around each node consisting of its k-neighborhood (e. g., k = 3). In ZRP, the distance and a node, all nodes within -hop distance from node belongs to the routing zone of node {/** what does this sentence mean? **/}. ZRP is formed by two sub-protocols, Intra-zone Routing Protocol (IARP) and Inter-zone Routing Protocol (IERP). IARP is a proactive routing protocol, which Intra-zone Routing Protocol (IARP) is used inside routing zones; on the other hand, IERP is and a reactive routing protocol that Inter-zone Routing Protocol (IERP) is used between routing zones respectively. A route to a destination within the local zone can be established from the proactively cached routing table of the source by IARP; therefore, if the source and destination is in the same zone, the packet can be delivered immediately. Most of the existing proactive routing algorithms can be used as the IARP for ZRP. For routes beyond the local zone, route discovery happens reactively. The source node sends a route requests to its border nodes, containing its own address, the destination address and a unique sequence number. Border nodes are nodes which are exactly the maximum number of hops to the defined local zone away from the source {/** what does this sentence mean? **/}. The border nodes check their local zone for the destination. If the requested node is not a member of this local zone, the node adds its own address to the route request packet and forwards the packet to its border nodes. If the destination node is a member of the local zone of the node, it sends a route reply on the reverse path back to the source. The source node uses the path saved in the route reply packet to send data packets to the destination. 1.1.2 Genetic Algorithm GAs [2] are adaptive heuristic search algorithms premised on the evolutionary ideas of natural selection and genetic. The basic concept of GAs is designed to simulate processes in natural system necessary for evolution, specifically those that follow the principles first laid down by Charles Darwin of survival of the fittest. As such they represent an intelligent exploitation of a random search within a defined search space to solve a problem. First pioneered by John Holland in the 60s, GAs have Genetic Algorithms has been widely studied, experimented and applied in many fields in engineering worlds. Not only does GAs provide alternative methods to solving problem, it consistently outperforms other traditional methods in most of the problems links. Many of the real world problems involved finding optimal parameters, which might prove difficult for traditional methods but ideal for GAs. However, because of its outstanding performance in optimization, GAs has have been wrongly regarded as a function optimizer. In fact, there are many ways to view genetic algorithms GAs. Perhaps most users come to GAs looking for a problem solver, but this is a restrictive view. 1.1.3 Estimation of Distribution Algorithms Estimation of Distribution Algorithms (EDAs) [6], sometimes called Probabilistic ModelBuilding Genetic Algorithms (PMBGA), are an outgrowth of genetic algorithms GAs. In a genetic algorithm GA, a population of candidate solutions to a problem is maintained in order to as part of the search for an optimum solution. This population is typically represented explicitly as an array of objects. Depending on the specifics {/** what does this term means? **/} of the GA, the objects might be bit strings, vectors of real numbers, LISP style S expressions or some custom representation. In an EDA, this explicit representation of the population is replaced with a probability distribution over the choices available at each position in the vector that represents a population member. Martin Pelikan started working in University of Illinois, in ILLIGAL lab, on estimation of distribution algorithm EDA in 1999. Since then, it is becoming popular and started to use as a qualified alternative to Genetic algorithm GA in applications and theory. 1.2 Related Work In recent past, a lot of attention has been shown by the research community to various issues related to ad hoc networks [15, 16] [15],[16]. Many protocols have been proposed for routing in such an environment. These protocols can broadly be classified into two types: proactive and reactive routing protocols. Proactive or table-driven protocols try to maintain routes to all the nodes in the network at all times by broadcasting routing updates in the network. Examples are DSDV [17], TBRPF, OLSR, WRP, STAR, and FSR. On the other hand, reactive or on-demand protocols attempt to find a route to the destination, only when the source has a packet to send to the destination. Examples are DSR [18], AODV, and TORA. Proactive protocols maintain the routing information of one node to the other using routing tables. Whenever there is a need for the route to the destination, it is readily available incurring minimum delay. But, at the same time, they may lead to a lot of wastage of the network resources if a majority of these available routes are never used. Reactive protocols are usually associated with less control traffic in a dynamic network; nodes have to wait until replies to the route queries are received. Also reactive protocols resort to frequent flooding of the network, which may cause network congestion. In between the above two extremes, there are the hybrid protocols. The ZPA Zone Routing Protocol (ZRP) [19] is a hybrid proactive / reactive protocol. It is a routing framework composed of the proactive IARP Intra-zone Routing Protocol (IARP) [20], reactive IERP Inter-zone Routing Protocol (IERP) [21], and the Border cast Resolution Protocol (BRP) [22]. ZRP is proved to work well compared to either table-driven protocols [23-26] or on-demand protocols [27, 28] [23-26],[27-28]. Genetic Algorithms GAs perform much better with rugged landscapes because of their population based approach spreading “probes” throughout the search space. In our work we preferred GA genetic algorithm as the optimization algorithm because of the confidence that it would work due to its robustness. Note that one other big advantage of GAs genetic algorithms is the ability to parallelize them on a large scale by spreading the evaluations across different machines. A large amount of work has been done on the application of GAs genetic algorithms or evolutionary algorithms to communications networks. Investigators have applied GAs to the shortest path routing problem [6], multicast routing problem [7-9]. In [10], the authors have tried to investigate the used of genetic algorithms GAs for automated selection of parameters like heartbeat interval, heartbeat points, score history size, up score threshold, down score threshold, routing algorithm and traffic max attempts for ad hoc networks. The authors of [11] have presented a genetic algorithmic approach to the shortest path routing problem. [12] Investigates using a GA genetic algorithm to create routing tables for an under water ad hoc networks. [13] examines the application of GAs genetic algorithms to dynamically optimize routing in MANETs. [14] Looks at how to use a GA genetic algorithm to cluster network nodes into subnetworks. 1.3 Objective of our Thesis Recently GA genetic algorithm has been used in [3][4][5], MANET mobile ad hoc network to find the optimized solution. This gives a better performance than the traditional zone routing protocol. EDAs Estimation of distribution algorithms are widely being accepted by the researcher around the world as an alternative for GAs Genetic algorithms. EDA is based on probabilistic model, thus converges faster than GA genetic algorithm. Hence, when we studied the Genetic Zone Routing Protocol (Known as when GA genetic algorithm is used to solve ZRP), we considered EDA as an alternative of GAs Genetic algorithms and found it very interesting for research {/** how? **/}. Our objectives of research can be summarized as follows:       To implement the traditional Zone Routing Protocol. To implement and compare the Genetic Zone Routing Protocol with the traditional Zone Routing Protocol. To survey the scope of using Estimation of Distribution Algorithms as an alternative of Genetic algorithm. To apply the Univariate Marginal Distribution Algorithm (UMDA) as an Estimation of Distribution Algorithm. To compare and analyze the performance of EDA and GA in ZRP To find the reasonable solution that stands for the comparison result of GA and EDA. As EDA is still a growing field of research and so far no one came out of trying to solve ZRP with EDA, we hope that this research will contribute a bit in analyzing the performance of EDA in a new application area, which is already solved by Genetic algorithm. The observations will surely be interesting one. Chapter 2 BACKGROUND 2.1 Zone Routing Protocol (ZRP) The Zone Routing Protocol, as its name implies, is based on the concept of zones. A routing zone is defined for each node separately, and the zones of neighboring nodes overlap. The routing zone has a radius r expressed in hops. The zone thus includes the nodes, whose distance from the node in question is at most r hops. An example routing zone is shown in Figure 1, where the routing zone of S includes the nodes A–I, but not K. In the illustrations, the radius is marked as a circle around the node in question. It should however be noted that the zone is defined in hops, not as a physical distance. Figure 1: Example routing zone with r=2 The nodes of a zone are divided into peripheral nodes and interior nodes. Peripheral nodes are nodes whose minimum distance to the central node is exactly equal to the zone radius r. The nodes whose minimum distance is less than r are interior nodes. In Figure 1, the nodes A–F are interior nodes, the nodes G–J are peripheral nodes and the node K is outside the routing zone. Note that node H can be reached by two paths, one with length 2 and one with length 3 hops. The node is however within the zone, since the shortest path is less than or equal to the zone radius. The number of nodes in the routing zone can be regulated by adjusting the transmission power of the nodes. Lowering the power reduces the number of nodes within direct reach and vice versa. The number of neighboring nodes should be sufficient to provide adequate reachability and redundancy. On the other hand, a too large coverage results in many zone members and the update traffic becomes excessive. Further, large transmission coverage adds to the probability of local contention. ZRP refers to the locally proactive routing component as the IntrA-zone Routing Protocol (IARP). The globally reactive routing component is named IntEr-zone Routing Protocol (IERP). IERP and IARP are not specific routing protocols. Instead, IARP is a family of limited-depth, proactive link-state routing protocols. IARP maintains routing information for nodes that are within the routing zone of the node. Correspondingly, IERP is a family of reactive routing protocols that offer enhanced route discovery and route maintenance services based on local connectivity monitored by IARP. The fact that the topology of the local zone of each node is known can be used to reduce traffic when global route discovery is needed. Instead of broadcasting packets, ZRP uses a concept called bordercasting. Bordercasting utilizes the topology information provided by IARP to direct query request to the border of the zone. The bordercast packet delivery service is provided by the Bordercast Resolution Protocol (BRP). BRP uses a map of an extended routing zone to construct bordercast trees for the query packets. Alternatively, it uses source routing based on the normal routing zone. By employing query control mechanisms, route requests can be directed away from areas of the network that already have been covered. In order to detect new neighbor nodes and link failures, the ZRP relies on a Neighbor Discovery Protocol (NDP) provided by the Media Access Control (MAC) layer. NDP transmits “HELLO” beacons at regular intervals. Upon receiving a beacon, the neighbor table is updated. Neighbors, for which no beacon has been received within a specified time, are removed from the table. If the MAC layer does not include a NDP, the functionality must be provided by IARP. The relationship between the components is illustrated in Figure 2. Route updates are triggered by NDP, which notifies IARP when the neighbor table is updated. IERP uses the routing table of IARP to respond to route queries. IERP forwards queries with BRP. BRP uses the routing table of IARP to guide route queries away from the query source. Detailed of ZRP can be found in [1]. Figure 2: ZRP Architecture 2.2 Genetic Algorithm (GA) A genetic algorithm is a type of searching algorithm. It searches a solution space for an optimal solution to a problem. The key characteristic of the genetic algorithm is how the searching is done. The algorithm creates a "population" of possible solutions to the problem and lets them "evolve" over multiple generations to find better and better solutions. The generic form of the genetic algorithm is found in Algorithm bellow. The items in bold in the algorithm are defined here. Algorithm: a General Genetic Algorithm BEGIN 1. Create a population of random candidate solutions named pop. 2. Until the algorithm termination conditions are met, do the following (each iteration is called a generation): (a) Create an empty population named new-pop. (b) While new-pop is not full, do the following: i. Select two individuals at random from pop so that individuals which are more fit are more likely to be selected. ii. Cross-over the two individuals to produce two new individuals. (c) Let each individual in new-pop have a random chance to mutate. (d) Replace pop with new-pop. 3. Select the individual from pop with the highest fitness as the solution to the problem. END The population is the collection of candidate solutions that we are considering during the course of the algorithm [5]. Over the generations of the algorithm, new members are "born" into the population, while others "die" out of the population. A single solution in the population is referred to as an individual. The fitness of an individual is a measure of how "good" the solution represented by the individual is. The better the solution, the higher the fitness - obviously, this is dependent on the problem to be solved. The selection process is analogous to the survival of the fittest in the natural world. Individuals are selected for "breeding" (or cross-over) based upon their fitness values-the fitter the individual, the more likely that individual will be able to reproduce [5]. The cross-over occurs by mingling the two solutions together to produce two new individuals. During each generation, there is a small chance for each individual to mutate, which will change the individual in some small way. To use a genetic algorithm, there are several questions that need to be answered {/** upon what criterion, you are claiming that the followings are to be answered? **/}:     How is an individual represented? How is an individual's fitness calculated? How are individuals selected for breeding? How are individuals crossed-over?    How are individuals mutated? How big should the population be? What are the "termination conditions"? Most of these questions have problem-specific answers. The last two, however, can be discussed in a more general way. The size of the population is highly variable. The larger the population, the more possible solutions there are, which means that there is more variation in the population. Variation means that it is more likely that good solutions will be created. Therefore, the population should be as large as possible. The limiting factor is, of course, the running time of the algorithm. The larger the population, the longer the algorithm takes to run. The algorithm of the General Genetic Algorithm presented above has a very vague end point the meaning of "until the termination conditions are met" is not immediately obvious. The reason for this is that there is no one way to end the algorithm. The simplest approach is to run the search for a set number of generations- the longer it will be run it, the better. Another approach is to end the algorithm after a certain number of generations pass with no improvement in the fitness of the best individual in the population. There are other possibilities as well. Detailed Deatailed of GA can be found in [2]. 2.3 Estimation of Distribution Algorithm (EDA) In EDAs the problem specific interactions among the variables of individuals are taken into consideration. In Evolutionary Computations the interactions are kept implicitly in mind whereas in EDAs the interrelations are expressed explicitly through the joint probability distribution associated with the individuals of variables selected at each generation. The probability distribution is calculated from a database of selected individuals of previous generation. The selections methods used in Genetic Algorithm may be used here. Then sampling this probability distribution generates offspring. Neither crossover nor mutation has been applied in EDAs. But the estimation of the joint probability distribution associated with the database containing the selected individuals is not an easy task. The following is a pseudo-code for EDA approach: Step 1: D0  Generate M individuals (the initial population) at random Step 2: Repeat steps 3-5 for l=1, 2, … until the stopping criteria met Dlse1   Step 3: method se Step 4: pl ( x)  p (x | Dl 1 )  Estimate the probability distribution of an Select N<=M individuals from Dl-1 according to selection individual being among the selected individuals Step 5: Dl  Sample M individuals (the new population) from pl(x) The easiest way to calculate the estimation of probability distribution is to consider all the variables in a problem as univariate. Then the joint probability distribution becomes the product of the marginal probabilities of n variables, i.e. pl (x)  i 1 p( xi ) n . Univariate Marginal Distribution Algorithm (UMDA) (Mühlenbein, 1998), Population Based Incremental Learning (PBIL) (Baluja, 1994) and Compact Genetic Algorithm (cGA) (Harik et al., 1998) consider no interaction among variables. In UMDAs the joint probability distribution is factorized as a product of independent univariate marginal distribution, which is estimated from marginal frequencies:  p (x )  l i N j 1  j ( X i  xi | Dlse1 )  N with  j ( X i  xi | Dlse1 )  1 if in the jth case of Dlse1 ,   2.3.1 Xi=xi ; =0, otherwise. Univariate Marginal Distribution Algorithm (UMDA) In UMDA it is assumed that is there is no interrelation among the variables of the problems. Hence the n-dimensional joint probability distribution is factorized as a product of n univariate and independent probability distribution. That is: pl (x)  p (x | Dlse1 )   p l ( xi )  i 1 n . distribution is estimated from marginal frequencies: Each univariate N j 1 marginal  p (x )  l i  j ( X i  xi | Dlse1 )  N with  j ( X i  xi | Dlse1 )  1 if in the jth case of Dlse1 , X =x   i i =0, otherwise. The pseudocode for UMDA is as follows: Step-1:D0Generate M individuals (the initial population) at random Step-2: Repeat steps 3-5 for l=1,2,… until stopping criteria is met Step-3: Dl-1 Select NM individuals from Dl-1 according to selection method Step-4: Estimate the joint probability distribution pl (x)  p (x | DlSe1 )   p l ( xi )  i 1 n Step-5: Dl Sample M individuals (the new population) from pl(x) 2.3.2 Convergence of UMDA by Laplace correction González et al. has shown that some instances with pl(x)>0 visits populations of D* which contains global optimum infinitely with probability 1 and if the selection is elitist, then UMDA may converge to a population that contains the global optimum. But the joint probability distribution of UMDA can be zero for some x; for example, when the Se selected individuals at the previous steps are such that  j ( X i  xi | Dl 1 )  0 for all j=1, 2, …, N. Hence pl(x)=0. So UMDA sometimes may not visit a global optimum (González et al.). To overcome the problems the way of calculating the probabilities should be changed. One possible solution Se l 1 is N j 1 to apply Laplace correction (Cestnik, 1990). Now p ( X i  xi | D  )  j ( X i  xi | DlSe1 )  1  N  ri where ri is the number of different values that variable Xi may take. Detailed of EDA can be found in [6]. Chapter 3 PROPOSED METHOD 3.1 Motivation We have learned found from our study in [3, 4, 5] [3][4][5] that, zone routing protocol is widely accepted and a popular routing mechanism in MANET mobile ad hoc network. We already know that, ad hoc network forms the temporary network without any support of infrastructure. So, there are possibilities of unwanted delay and lack of reliability in the network. Again, if the number of nodes grows, the linear search will become costly and the complexity will become high. In case of large number of nodes, a random search will be beneficial where the worst case will equal the linear search. Here GA genetic algorithm may play plays an important role in optimizing the search. This is because GA genetic algorithm calculates the fitness of each population and generates a better population using crossover and mutation. So, the chance of getting good solutions increases dramatically. But, there are cases that GAs genetic algorithms will never converge to the optimal point. And in some of the cases, GA genetic algorithm takes longer time than expected. This may happen due to poor encoding structure or the less effective fitness function. However, the main reason would be the generation of a poor solution from two good solutions. This is the point where EDAs estimation of distribution algorithms works better than GAs genetic algorithms. Strictly speaking, GA and EDA are same, apart form the crossover and mutation. There is nothing called crossover and mutation in EDA. Instead they use probabilistic model for generating new population. This guarantees the better generation of population than earlier generation. So, EDA converges faster, even if there is no feasible solution. So, the chances are high that in case of ZRP Zone Routing Protocol, EDA will perform better than GA which currently performs better than the traditional method. This requires a lot of research and analysis to support the viewpoint. So, in our standpoint, it is an obvious and interesting research area, which can contribute a lot in a new application area of EDA. 3.2 Research Focus There are various types of EDA [6] both in discrete and continuous domain. We will consider the discrete domain in our case. Specifically, we want to use UMDA [1] to apply EDA in ZRP. UMDA is the simplest of all EDA and use univariate distribution to generate new population. Therefore, the implementation will be easier and this can be compared to the result of GA in ZRP. We want to concentrate on a good encoding strategy that will be same for both GA and EDA. The additional thing used in GA will be one-point crossover and mutation, whereas in EDA, it will be the probability model. Then we would like to compare and analyze the result of both the method. We want to focus more on logical and mathematical reasoning of performance of EDA and GA rather than using various EDA on same method. 3.3 Research Methodology Our main source of information will be the published papers on ZRP, GA and EDA. We want to go through in detail study of these papers, and find the best possible scenario for our thesis. Existing implementation will give us an overview of practical knowledge of ad hoc network. However, as we are unable to implement the hardware details of ad hoc network, we want to concentrate on the simulation. Besides, our research is based on theoretical aspect of mathematics. So, we can hopefully come with a successful thesis. 3.4 Proposed Method Our proposed method can be summarized as follows:         To find a good encoding strategy that will represent the ad hoc network. This encoding will give us the vector which we will try to convert into a binary representation. Then the representation will act as a chromosome of the population. To randomly generate the initial population. To calculate the fitness value for each chromosome. Use the same fitness function for both GA and EDA. To perform crossover and mutation for GA to generate new population. To perform Probabilistic model for EDA to generate new population. Select the subpopulation with elitism and without elitism. Continue until the result converges. {/** These are the objectives, not the proposed method **/} 3.5 Implementation Outline To be realistic, we will try to use as large network as possible that can give us a solution which will be usable in practical purpose. We will use either of MATLAB or C++ to implement the proposed method. Existing dataset of ad hoc network can also be used. So, we will try to use any popular existing network. That will make our research wonderful in terms of practical use. The main problem will be occurred when the number of chromosome length and population length increase. This makes the calculation huge and time consuming. We will try to optimize the parameters as close as possible with the existing network. Chapter 4 SUMMERY 4.1 Discussion For the optimized performance, we need a good encoding strategy that will not be computationally complex. We could not find a good strategy so far. A great deal of work has to be done to find the data set and the benchmark for the problem. We did not calculate the cost analysis and computation difficulties so far. These things are to be solved before implementation. We need to find a good selection mechanism that will cope with ad hoc network. 4.2 Conclusion and Future work We have a lot of analysis to do. With the help of our supervisor, we will be able to solve the problems of the research area. It demands a great deal of time to complete the entire project. We can also increase the scope of our thesis in future to cope up with the multicast routing in ad hoc network with dynamic behavior, is the current work is done successfully.