A New Parallel Algorithm to Solve the Near-Shortest-Path Problem on Raster Graphs

The Near Shortest Path algorithm (Carlyle and Wood 2005) has been identified as being effective at generating sets of good route alternatives for developing new infrastructure. While the algorithm itself is faster than other shortest path set approaches such as solving the k-th shortest path problem, the solution set size and computation time grows exponentially as the problem size increases and requires the use of high-performance parallel computing to solve real world corridor location problems. We identified a new breadth-first-search parallelization of the Near Shortest Path algorithm, although its efficiency was limited by large discrepancies in workloads from each processing thread. In an effort to more equally distribute work, we defined a metric that can be used to predict workloads from different parts of the breadth first search tree reducing the overall variability of workload between threads. This resulted in much improved algorithm performance and parallel computing efficiencies. Future work should focus on refining this new approach, and developing guidelines for implementing this method over a variety of datasets.