May 2005, Volume 2, No.5 (Serial No.

6)

Journal of Communication and Computer, ISSN1548-7709, USA

Multicast Routing Optimization Algorithm with Bandwidth and Delay Constraints Based on GA*
Hua Chen, Baolin Sun
(Department of Mathematics and Physics, Wuhan University of Science and Engineering, Wuhan 430073, China) Abstract: Most of the multimedia applications require strict QoS guarantee during the communication between a single source and multiple destinations. This gives rise to the need for an efficient QoS multicast routing strategy. Determination of such QoS-based optimal multicast routes basically leads to a multi-objective optimization problem, which is computationally intractable in polynomial time due to the uncertainty of resources in networks. This paper proposes a new multicast routing optimization algorithm based on Genetic Algorithms, which find the low-cost multicasting tree with bandwidth and delay constraints. The simulation results show that the proposed algorithm is able to find a better solution, fast convergence speed and high reliability. It can meet the real-time requirement in multimedia communication networks. The scalability and the performance of the algorithm with increasing number of network nodes are also quite encouraged. Key words: QoS; Multicast Routing; Genetic Algorithms; Routing Optimization*

1. Introduction
Multicast services have been used by various continuous media applications. For example, the multicast backbone (Mbone) of the Internet has been
Acknowledgements: This work is supported by National Natural Science Foundation of China (No. 60172035, 90304018), NSF of Hubei Province (No. 2004ABA014), and Teaching Research Project of Higher Educational Institutions of Hubei Province (No. 20040231). Hua Chen received MS degree in mathematics of Huazhong University of Science and Technology. She is currently a lecturer at the Department of Mathematics and Physics at Wuhan University of Science and Engineering. Her research interests include genetic algorithms and distributed computing. Baolin Sun, PhD candidate in computer science of Wuhan University of Technology, associate professor at the Department of Mathematics and Physics at Wuhan University of Science and Engineering. He has published one book and over 40 journal papers. His research interests include QoS multicast routing, genetic algorithms, distributed computing and protocol engineering. E-mail: sun0163@163.com.
*

used to transport real time audio/video for news, entertainment, video conferencing, and distance learning[1-3, 5-7,10]. The provision of Quality-of-Service (QoS) guarantees is of utmost importance for the development of the multicast services. Multicast routing has continued to be a very important research issue in the areas of network and distributed systems. It attracts the interests of many people. QoS multicast routing relies on state parameters specifying resource availability at network nodes or links, and uses them to find paths with enough free resources [1-3, 11]. In turn, the successful routing of new flows together with the termination of existing ones, induces constant changes in the amount of available resources. These must be communicated back to QoS multicast routing. Unfortunately, communicating such changes in a timely fashion is expensive and, at times, not even feasible [2,4,5]. In this paper, we design a multicast routing optimization algorithm with bandwidth and delay constraints based on Genetic Algorithms which can be suitable to the networks with uncertain parameters. The focus is on determining multicast routes from a source to a set of destinations with strict end-to-end delay requirements and minimum bandwidth available. The goal of this paper is to develop an algorithm to find out multicast routes with bandwidth and delay constraints by simultaneously optimizing end-to-end delay and bandwidth provided for guaranteed QoS. The rest of the paper is organized as follows. Section 2 introduces a network model. Section 3 presents the multicast routing optimization algorithm
63

Multicast Routing Optimization Algorithm with Bandwidth and Delay Constraints Based on GA

based on Genetic Algorithms. Some simulation results are provided in Section 4. The paper conclusion and future research are presented in Section 5.

3. Genetic Algorithms
3.1 Encoding Representation The chromosomes of genetic algorithms are composed of a series of integral queuing and the encoding method is based on routing representation, which is the most natural and simplest representing method. Given a source node n0 and destination nodes set U={u1, u2, , um}, a chromosome can be represented by a string of integers with length m. The chromosome of genetic algorithms is composed of a series of integral queuing with length m, and the gene of genetic algorithms is the path in path set { Pi1 , , [10,11] between n0 and ui, where, Pi j is the Pi j , , Pi l } j-th path of destination node ui, l denotes the path number between n0 and ui. Each chromosome in population denotes a multicast tree. Obviously, a chromosome represents a candidate solution for the multicast routing problem since it guarantees a path between the source node and any of the destination nodes. Since there are so many paths between node n0 and ui, so that the encoding space of chromosomes possibly becomes larger, which decreases the convergence of solution. Now for each destination node d U, by the k-th the shortest route algorithm, the encoding space can be improved by finding out all routes that satisfy bandwidth constraint from source node n0 to destination node d U and composing routes set as candidate routes set of genetic algorithms encoding space. Assume that Ui is the set of destination node ui which satisfies bandwidth constraint, then,Ui={ Pi1 , , Pi j , ..., Pi k }, k = l, where, Pi j denotes the j-th route which satisfies bandwidth constraint of destination node ui. Choose arbitrarily a route from each route set Ui respectively, and compose of the initial population of chromosomes.

2. Network Model
A network is usually represented as a weighted digraph G = (N, E), where N denotes the set of nodes and E denotes the set of communication links connecting the nodes. |N| and |E| denote the number of nodes and links in the network respectively [1-3, 6, 10]. We consider the multicast routing problem with bandwidth and delay constraints from one source node to multi-destination nodes. Let M= {n0, u1, u2, , um} N be a set of form source to destination nodes of the multicast tree. Where n0 is source node, and U= {u1, u2 um} denotes a set of destination nodes. Multicast tree T= (NT, ET), where NT N, ET E, there exists the path PT(n0, d) from source node n0 to each destination node d U in T [1-3, 10]. Definition 1: The cost of multicast tree T is: C(T)= eET C (e) . Definition 2: The bandwidth of multicast tree T is the minimum value of link bandwidth in the path from source node n0 to each destination node d U. i.e. B(T) = min(B(e), e ET). Definition 3: The delay of multicast tree T is the maximum value of delay in the path from source node n0 to each destination node d U. i.e. D(T) = max (ePT ( n0 ,d ) D(e), d U ) . Definition 4: Assume the minimum bandwidth constraint of multicast tree is B, and the maximum delay constraint id is D, given a multicast demand R, then, the problem of bandwidth-delay constrained multicast routing is to find a multicast tree T, satisfying: (1) Bandwidth constraint: B(T) = B. (2) Delay constraint: D(T) = D. Suppose S(R) is the set, S(R) satisfies the conditions above, then, the multicast tree T which we find is: C(T) = min (C (Ts), Ts S(R))

64

Multicast Routing Optimization Algorithm with Bandwidth and Delay Constraints Based on GA

Chromosome g1 g2 gi gi=1 Routes set of satisfying bandwidth constraint between node n0 and the ui Route no. Route path 0 n0 ui 1 n0 ui k n0 ui Fig.1 Representation of Chromosomes gm

3.2 Fitness Sharing Function Genetic algorithms use rarely the outer information in evolution searching. Instead they use fitness function as basis and the values of every individual in population for searching. As a result, fitness function affects directly the convergent speed of genetic algorithm and whether the global optimum is found. The fitness value of a chromosome is the value of the fitness function for the solution (e.g., a multicast tree) represented by the chromosome. Given an initial population H={h1, h2, , hp}, the fitness value of each chromosome is computed as follows. Let C(hi) be the sum of the costs of the links of the graph represented by the chromosome hi and C(L) be the sum of the costs of all the links in the network. The fitness value of the chromosome hi, F(hi), is given by F(hi) = 1- C(hi)/C(L) ( 0 = F(hi) < 1 ) 3.3 Selection Operations Selection operation is used to certain or crossover individuals and selected individual can produce many sub-individuals. Selection operation has two procedures: firstly, computing fitness value; secondly, queuing it from the biggest to the smallest, namely, F(h1)=F(h2)= =F(hp). Then, the max fitness value is the best individual, selecting the best individual as father-individual. The selection probability of each individual is proportional to its fitness value, and the selected probability is higher when the individual fitness value is bigger. If the same chromosomes have been got, only one chromosome exists. The rest

chromosomes can be canceled. 3.4 Crossover and Mutation Operations As the algorithm executes, at each iteration we get a set of non-dominated strings whose fitness values represent the Pareto-optimal solutions for that iteration. The crossover and mutation operations are the same as normal genetic algorithms. But, it must be made sure that these operations must not produce any illegal paths. A close look into the structure of the chromosome in Fig.1 reveals that these genetic operations cannot be performed on any arbitrary gene (network nodes), as that can gives birth to some paths which do not exist at all. Both the crossover and mutation operations can only be performed at the end of an existing path. Giving an equal probability to all such possible crossover and mutation points, we randomly select one such point. The crossover operation is performed by swapping the portion of the two consecutive chromosomes after the particular selected point. In case of mutation we just replace the part of the chromosome after the mutation point by a corresponding part of any other valid chromosome. To combine the good strings and simultaneously preserve the effective ones we have taken the probability of cross over as 0.4 and that of mutation as 0.02 [7-12].

4. Algorithm Analysis and Simulations Results

4.1 Analysis of Convergence Theorem 5: The genetic algorithms proposed in this paper converge to the global optimal solution. Proof: The genetic algorithm has following merits: (1) The method which uses the candidate routing set from source node to each destination node, makes the searching speed faster, and the whole process could be done in shorter time; (2) Selecting by proportion and pertaining the optimal individual before selection; (3) Changeable length chromosome encoding method which based on routing expression is used; (4) Crossover probability between [0,1]; (5) Mutation probability

65

Multicast Routing Optimization Algorithm with Bandwidth and Delay Constraints Based on GA

between [0,1], by theorem 2.7 in reference[12], has crossover probability between [0,1], mutation probability between (0,1), and at the same time, the genetic algorithm obtained according to the method proposed above can converge to the global optimal solution. Hence, the genetic algorithm proposed in this paper can converge to the global optimal solution. 4.2 Computing Complexity The genetic algorithm is a parallel algorithm, and the searching speed is very fast, in which the whole process can be done in short time. Thus, the time complexity of algorithm in this paper is decided by the time complexity of reference [4], which is O(|E|+|N|log|N|+k). 4.3 Simulations Results Adopting the network topology structure as shown in Fig.2 in simulation experiment, the characteristic of links or sides in figure can be represented by a triple-group (Bij, Dij, Cij), given its value randomly. Assuming the source node n0 is node 1, destination node set U={4, 5, 7, 8}, the smallest bandwidth constraint B=10, by the algorithm for finding the k-th shortest paths in reference [4], we can find the candidate route set from source node 1 to each destination node, as shown in table 1.
Table 1 The Candidate Path Set from Source Node to Each Destination Node Destination node 4 5 7 8 Candidate path set {1,2,4},{1,3,4},{1,5,6,4} {1,5} {1,3,4,6,7},{1,5,6,7},{1,2,8,7},{1,2,4,6,7} {1,5,6,7,8},{1,3,4,6,7,8},{1,2,4,6,7,8},{1,2,8}

obvious when network scale is bigger, and bandwidth constraint is amplified. Repeat the simulation with increasing number of network nodes and the efficiency of our algorithm. As the network becomes highly condensed, our algorithm exhibits a more linear and stable pattern than existing scalar optimization algorithm. This approximate linearity of the curve in Fig.5 corroborates the scalability of the algorithm. Fig.6 shows the change in average network delay with the number of generations. It is observed that the network delay decreases with the increasing in the number of generation.
(15,3,6) 1 (13,3,4) 5 (12,2,6) 6 (8,3,6) (10,3,6) (10,2,4) 7 (12,2,6) (10,1,5) (15,2,8) 3 (12,3,4) 4 (13,2,6) (9,3,4) 8 2

(9,1,3)

Fig.2 Network Topology Structure

(15,3,6) 1 (13,3,4) 5 (12,2,6) 6 (8,3,6) (10,3,6) (10,1,5) (15,2,8) 3 (12,3,4) (13,2,6) 4 (9,3,4) 8 2

(9,1,3) (10,2,4) 7 (12,2,6)

Fig.3 Genetic Algorithm Generates Multicast Tree (B=10, D=7)

32 30 Reference [10] s algorithm Proposed algorithm

In this genetic algorithm, let crossover probability be 0.9, mutation probability be 0.2. When bandwidth constraint B=10, delay constraint D=7, generated multicast tree is as shown in Fig.3. We compare this algorithm with the one in reference [10].Fig.4 denotes the convergence comparison of cost with operation algebra of multicast tree generated by the two algorithms. This algorithm can quickly generate the optimal solution, furthermore, its advantage is more

Cost

28 26 24 10 40 80 120 160 200 240 280

Generation Fig.4 The Effect of the Cost with Genetic Generation

66

Multicast Routing Optimization Algorithm with Bandwidth and Delay Constraints Based on GA

Num. of generation

350 300 250 200 150 100 50 10 20 30 40 50 60 70 80 90 100 Reference [10] s algorithm Proposed algorithm

Number of nodes Fig.5 Performance of the Algorithm with Increasing Number of Nodes
20

environment, high dynamic and the requirement of network structure of QoS multicast routing. QoS multicast routing is the foreland research project in networks and information technology. While interests of many people have got better results in single constraint (especially delay constraint) multicast routing, the results based on multi-constraint QoS multicast routing are not satisfying. This algorithm can expand to multi-constraint s QoS multicast routing problem based on this algorithm; the chromosome of fitness function alone can be changed and the delay constraint can be improved so as to make the algorithm widely applied.
References: [1] Baolin Sun, Layuan Li, A QoS Based Multicast Routing Protocol in Ad Hoc Networks, Chinese Journal of Computers, Vol. 27, No.10, October 2004: pp.1402-1407(in Chinese). [2] Baolin Sun and Layuan Li, Multiple Constraints-Based QoS Multicast Routing: Model and Algorithms, Journal of Systems Engineering and Electronics, Vol. 15, No. 4, 2004. [3] Hua Chen and Baolin Sun, QoS-Based Multicast Routing Optimization Algorithms for Wireless Networks, 2004 International Symposium on Distributed Computing and Applications to Business, Engineering and Science (DCABES 2004), September 13-16, Wuhan, China, 2004: pp. 281-285. [4] D. Eppstein, Finding the k Shortest Paths, SIAM J. Computing, Vol.28, No.2, 1998: pp.652-673. [5] J. J. Wu, R.-H. Hwang, Multicast Routing with Multiple Constraints, Information Sciences, Vol. 124, 2000: pp. 29-57. [6] Z. Xiawei, C. Changjia, Z. Gang, A Genetic Algorithm for Multicasting Routing Problem, International Conference Communication Technology Proceedings, WCC-ICCT 2000, 2000: pp. 1248-1253. [7] Zhang Q, Lenug Y. W, An Orthogonal Genetic Algorithm for Multimedia Multicast Routing, IEEE Trans Evolutionary Computation, Vol. 3, 1999: pp. 53-62. [8] M. Munemoto, Y. Takai, and Y. Sato, A Migration Scheme for the Genetic Adaptive Routing Algorithm, IEEE International Conference on Systems, Man, and Cybernetics, 1998: pp. 2774-2779. [9] J. Inagaki, M. Haseyama, H. Kitajima, A Genetic Algorithm for Determining Multiple Routes and Its Applications, Proceedings of IEEE International Symposium on Circuits and Systems, 1999: pp. 137-140. [10] R. -H. Hwang, W. -Y. Do and S. -C. Yang, Multicast Routing Based on Genetic Algorithms, Journal of Information Science and Engineering, Vol. 16, 2000: pp. 885-901. [11] N. Shimamoto, A. Hiramatu, K. Yamasaki, A Dynamic Routing Control Based on a Genetic Algorithm, 1993 IEEE International Conference on Neural Network, 1993: pp. 1123-1128. [12] Guoliang Chen et al, Genetic Algorithm and Its Application, Beijing: People s Posts and Telecommunications Press, 1996(in Chinese).

Delay(ms)

15 10 5 0 0 20 40 60 80 100 120 140 160 180 200

Generation Fig.6 Average Network Delay vs. Generation

5. Conclusions
This paper proposes a QoS multicast routing model and optimization algorithm based on bandwidth and delay constraints and gives the heuristic genetic algorithm of minimum-costs QoS multicast tree and bandwidth-delay constraint. This algorithm has merits as follows: (1) Algorithm is used to find out the k-th shortest path in reference [4] to pretreat, to construct candidate route set, to decrease searching space of algorithm efficiently, and to increase searching efficiency of algorithm marvelously; (2) Encoding operation based on the tree structure encoding of route simplified is by omitting complicated encodingdecoding process; (3) Guaranteed and speeded searching ability of the optimal solution and the global convergence of solution by heuristic crossover and mutation operation can be achieved; (4) The time complexity of the algorithm is O(|E|+|N|log|N|+k). Experiment represents that its convergent speed is fast and reliability. Especially in large network, the algorithm can greatly decrease routing computation time, satisfying the topology structure of real-time communication

(Editors: Wei Zhou, Ivan, Nory)

67