Mate, A Unified Model For Communication-Tolerant Scientific Applications

We present MATE, a model for developing communication-tolerant scientific applications. MATE employs a combination of mechanisms to reduce or hide the cost of network and intra-node data movement. While previous approaches have been proposed to reduce both sources of communication overhead separately, the contribution of MATE is demonstrating the symbiotic effect of reducing both forms of data movement taken together. Furthermore, MATE provides these benefits within a single unified model, as opposed to hybrid (e.g., MPI+X) approaches. We demonstrate MATE's effectiveness in reducing the cost of communication in three scientific computing motifs on up to 32k cores of the NERSC Cori Phase I supercomputer.