We discuss the data-parallel implementation of the epsilon-relaxation algorithm for the solution of min-cost network flow problems. For transportation problems a Gauss-Seidel variant is implemented on the two-dimensio...
详细信息
We discuss the data-parallel implementation of the epsilon-relaxation algorithm for the solution of min-cost network flow problems. For transportation problems a Gauss-Seidel variant is implemented on the two-dimensional communication grid of a massively parallel Connection Machine CM-2. For arbitrary network topologies - i.e. transhipment problems - we implement a Jacobi algorithm using the hypercube communication network. A new parallel flow-calculation procedure increases the level of parallelism at every iteration of the algorithm and improves its performance. Extensive computational results are reported with network problems with up to 1 million arcs.
暂无评论