Heterogeneous Multiconstraint Application Partitioner (HMAP)

In this article we propose a novel framework -- Heterogeneous Multiconstraint Application Partitioner (HMAP) for exploiting parallelism on heterogeneous High performance computing (HPC) architectures. Given a heterogeneous HPC cluster with varying compute units, communication constraints and topology, HMAP framework can be utilized for partitioning applications exhibiting task and data parallelism resulting in increased performance. The challenge lies in the fact that heterogeneous compute clusters consist of processing elements exhibiting different compute speeds, vector lengths, and communication bandwidths, which all need to be considered when partitioning the application and associated data. We tackle this problem using a staged graph partitioning approach. Experimental evaluation on a variety of different heterogeneous HPC clusters and applications show that our framework can exploit parallelism resulting in more than 3× speedup over current state of the art partitioning technique. HMAP framework finishes within seconds even for architectures with 100's of processing elements, which makes our algorithm suitable for exploring parallelism potential.