TY - GEN
T1 - Improving MPI independent write performance using a two-stage write-behind buffering method
AU - Liao, Wei Keng
AU - Ching, Avery
AU - Coloma, Kenin
AU - Choudhary, Alok
AU - Kandemir, Mahmut
PY - 2007
Y1 - 2007
N2 - Many large-scale production applications often have very long executions times and require periodic data check-points in order to save the state of the computation for program restart and/or tracing application progress. These write-only operations often dominate the overall application runtime, which makes them a good optimization target. Existing approaches for write-behind data buffering at the MPI I/O level have been proposed, but challenges still exist for addressing system-level I/O issues. We propose a two-stage write-behind buffering scheme for handing checkpoint operations. The first-stage of buffering accumulates write data for better network utilization and the second-stage of buffering enables the alignment for the write requests to the file stripe boundaries. Aligned I/O requests avoid file lock contention that can seriously degrade I/O performance. We present our performance evaluation using BTIO benchmarks on both GPFS and Lustre file systems. With the two-stage buffering, the performance of BTIO through MPI independent I/O is significantly improved and even surpasses that of collective I/O.
AB - Many large-scale production applications often have very long executions times and require periodic data check-points in order to save the state of the computation for program restart and/or tracing application progress. These write-only operations often dominate the overall application runtime, which makes them a good optimization target. Existing approaches for write-behind data buffering at the MPI I/O level have been proposed, but challenges still exist for addressing system-level I/O issues. We propose a two-stage write-behind buffering scheme for handing checkpoint operations. The first-stage of buffering accumulates write data for better network utilization and the second-stage of buffering enables the alignment for the write requests to the file stripe boundaries. Aligned I/O requests avoid file lock contention that can seriously degrade I/O performance. We present our performance evaluation using BTIO benchmarks on both GPFS and Lustre file systems. With the two-stage buffering, the performance of BTIO through MPI independent I/O is significantly improved and even surpasses that of collective I/O.
UR - http://www.scopus.com/inward/record.url?scp=34548738899&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=34548738899&partnerID=8YFLogxK
U2 - 10.1109/IPDPS.2007.370485
DO - 10.1109/IPDPS.2007.370485
M3 - Conference contribution
AN - SCOPUS:34548738899
SN - 1424409101
SN - 9781424409105
T3 - Proceedings - 21st International Parallel and Distributed Processing Symposium, IPDPS 2007; Abstracts and CD-ROM
BT - Proceedings - 21st International Parallel and Distributed Processing Symposium, IPDPS 2007; Abstracts and CD-ROM
T2 - 21st International Parallel and Distributed Processing Symposium, IPDPS 2007
Y2 - 26 March 2007 through 30 March 2007
ER -