Optimizing maximum shared risk link group disjoint path algorithm using NVIDIA CUDA heterogeneous parallel programming platform

Network availability is an essential feature of an optical telecommunication network. Should a failure of a network component occur, be it a link or a component inside a node, network control plane must be able to detect the failure and reroute the traffic using spare components until a repair is done. Shared risk link groups (SRLGs) are used to describe a situation where seemingly unrelated logical failures happen due to a single physical failure. For example, two or more links might share a bridge crossing; should a failure happen, all of them will be damaged. Routing algorithms were proposed to ensure working and spare paths of a connection in a network are SRLG-disjoint to avoid such common cause failures. However, complete SRLG-disjointness of working and spare path is not always possible due to limited number of links or limited capacity available in the network, so maximum SRLG-disjoint paths algorithm is taken instead. Maximum SRLG-disjoint path problem is in general NP-hard. In terms of solution quality greedy algorithms for maximum SRLG-disjoint path problem are as good as more complicated heuristics. To improve the performance of maximum SRLG-disjoint path greedy algorithm, it was implemented using NVIDIA CUDA heterogeneous parallel programming platform and executed on graphics processing unit. The implementation of maximum SRLG-disjoint path algorithm on GPU increases performance significantly compared to implementation utilizing only CPU, especially in simulations of large networks.