Scalable Compression and Replay of Communication Traces in Massively Parallel Environments

Profiling;Blue Gene L;HPC;MPI;communication;compression;scalable

Noeth, Michael James ; Dr. Tao Xie, Committee Member,Dr. Xiaosong Ma, Committee Member,Dr. Frank Mueller, Committee Chair,Noeth, Michael James ; Dr. Tao Xie ; Committee Member ; Dr. Xiaosong Ma ; Committee Member ; Dr. Frank Mueller ; Committee Chair

University:North Carolina State University

关键词: Profiling;    Blue Gene L;    HPC;    MPI;    communication;    compression;    scalable;   

Others  :  https://repository.lib.ncsu.edu/bitstream/handle/1840.16/1116/etd.pdf?sequence=1&isAllowed=y

美国|英语

来源: null

【 摘 要 】

Characterizing the communication behavior of large-scale applications isa difficult and costly task due to code and system complexity as well asthe time to execute such codes.An alternative to running actual codes is to gather their communication traces and then replay them, whichfacilitates application tuning and future procurements.While pastapproaches lacked lossless scalable trace collection, we contribute anapproach that provides near constant-size communication tracesregardless of the number of nodes while preserving structural information.We introduce intra- and inter-node compression techniques of MPI events and present results of our implementation.Given this novel capability, we discuss its impact on communication tuning andbeyond.