Experiences with Parallel Multi-threaded Network Maximum Flow Algorithm

The problem of computing the maximum flow problem on capacitated networks arises in many application areas. In the
area of heterogeneous computing, it arises in job or process scheduling when allocations of resources to jobs/processes
need to be tuned. The maximum flow solver is dicult to parallelize. Highly optimized sequential version of maximum flow solvers such as those by Goldberg exists. This work describes how some of the concurrency problems are resolved
in our existing Pmaxflow (https://code.google.com/p/pmaxflow ) solver. Pmaxflow employs a parallel pre-flow push
algorithm to compute the maximum flow. Results of various tests that compare Goldberg's sequential solvers and
Pmaxflow on a NUMA shared memory computer are presented.