Efficient Implementation of MPI-3 RMA over OpenFabrics Interfaces

Hajime Fujita,Chongxiao Cao,Sayantan Sur,Charles Archer, Erik Paulson,Maria Garzaran

Parallel Computing（2019）

引用 2|浏览76

暂无评分

摘要

•We developed a new MPI RMA implementation as a part of MPICH-OFI.•MPICH-OFI is based on MPICH and the CH4 device and uses OpenFabrics Intefaces.•MPICH-OFI achieves more than 3x better put/get latency and bandwidth against MPICH CH3.•10% better latency against Open MPI and MVAPICH2.•More than 1.7x bandwidth over MVAPICH2 for small messages ( ≤ 4KB).

查看译文

关键词

Message Passing Interface,MPI,Remote Memory Access (RMA),One-sided Communications,MPICH-OFI,OpenFabrics Intefaces (OFI)

AI 理解论文

溯源树

样例

生成溯源树，研究论文发展脉络

Chat Paper

正在生成论文摘要