Scalable Hybrid Implementation of Graph Coloring Using MPI and OpenMP

Graph coloring algorithms are commonly used in large scientific parallel computing either for identifying parallelism or as a tool to reduce computation, such as compressing Hessian matrices. Large scientific computations are nowadays either run on commodity clusters or on large computing platforms. In both cases, the current target platform is hierarchical with distributed memory at the node level and shared memory at the processor level. In this paper, we present a novel hybrid graph coloring algorithm and discuss how to obtain the best performance on such systems from algorithmic, system and engineering perspectives.