Communication-hiding programming for clusters with multi-coprocessor nodes

摘要

some light on the upcoming software challenge, this paper investigates the particular topic of programmingrnclusters that have multiple Xeon Phi coprocessors in each compute node. A new offload approach is consideredrnfor intra-node communication, which combines Intel’s APIs of coprocessor offload infrastructure (COI)rnand symmetric communication interface (SCIF) for achieving low latency. While the conventional pragmabasedrnoffload approach allows simpler programming, the COI-SCIF approach has three advantages in (1)rnlower overhead associated with launching offloaded code, (2) higher data transfer bandwidths, and (3) morernadvanced asynchrony between computation and data movement. The low-level COI-SCIF approach is alsornshown to have benefits over the MPI-OpenMP counterpart, which belongs to the symmetric usage mode.rnMoreover, a hybird programming strategy based on COI-SCIF is presented for joining the computationalrnforce of all CPUs and coprocessors, while realizing communication hiding. All the programming approachesrnare tested by a real-world 3D application, for which the COI-SCIF-based approach shows a performancernadvantage on Tianhe-2.