Designing a distributed algorithm for bandwidth allocation with a genetic algorithm

When stream communications which requires a certain bandwidth during a certain time is to be realized on an existing network, a bandwidth allocation algorithm that efficiently allocates network resources is crucial to satisfying user service requirements. In the optimization of allocation in a small-scale network, the most efficient approach is to have one node manage network information and perform allocation in a concentrated scheme. However, in bandwidth allocation the computation time required for optimization increases in proportion to the network scale, and distribution of the algorithm becomes necessary in the implementation. Another advantage of the distributed scheme is that allocations can be continued even if a fault arises in the network, which improves the fault tolerance of the algorithm. This paper investigates distribution of the bandwidth allocation algorithm by using a genetic algorithm. An improved algorithm is proposed which can cope with the fault when it occurs in the network by the reallocation of all communications. The proposed algorithm includes both local optimization to optimize the allocation at each node to reduce the computation time, and global optimization to optimize the whole network by the tournament method while performing local optimization in each node. Simulation experiments comparing the performance of the concentrated algorithm and the proposed algorithm are described. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(3): 37–45, 2004; Published online in Wiley InterScience (). DOI 10.1002sscj.10350