Performance Evaluation of Brute Force Techniques for Routing and Spectrum Assignment in Elastic Optical Network Using MPI and CUDA

In Elastic Optical Networks, Routing and Spectrum Assignment algorithms find a routing table and assign spectrum slots for each traffic demand such that the consumed spectrum on each optical link is minimized. This work deploys brute force algorithm to investigate all possible routing tables and select best candidates. Those candidates are then evaluated using the Spectrum Allocation algorithm to select the optimum spectrum utilization on the links. Routing table selection process depends on a scoring function that can be computed in linear time. Starting with an initial routing table (depending on shortest path algorithm) as an initial seed, the Selection criteria of the routing table decide whether to keep the initial seed or replace it with the one that can lead to a much more better solution. This in turn will eliminate the need to solve the spectrum assignment problem for each routing table. Consequently, the time needed to solve the problem will be dramatically reduced. Parallel computing paradigms are used to implement the brute force algorithm on parallel architectures to speed up the processing time. Performance comparison between MPI implementation and CUDA implementation is presented which shows that single GPU can perform better than high-end set of servers.