Y-lib: a user level library to increase the performance of MPI-IO in a lustre file system environment Export

@inproceedings{citeulike:4921049, abstract = {It is widely known that MPI-IO performs poorly in a Lustre file system environment, although the reasons for such performance are currently not well understood. The research presented in this paper strongly supports our hypothesis that MPI-IO performs poorly in this environment because of the fundamental assumptions upon which most parallel I/O optimizations are based. In particular, it is almost universally believed that parallel I/O performance is optimized when aggregator processes perform large, contiguous I/O operations in parallel. Our research shows that this approach generally provides the worst performance in a Lustre environment, and that the best performance is often obtained when the aggregator processes perform a large number of small, non-contiguous I/O operations.}, address = {New York, NY, USA}, author = {Dickens, Phillip M. and Logan, Jeremy}, booktitle = {HPDC '09: Proceedings of the 18th ACM international symposium on High performance distributed computing}, citeulike-article-id = {4921049}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1551609.1551617}, citeulike-linkout-1 = {http://dx.doi.org/10.1145/1551609.1551617}, doi = {10.1145/1551609.1551617}, isbn = {978-1-60558-587-1}, keywords = {concurrency, multicore, scalability, storage}, location = {Garching, Germany}, pages = {31--38}, posted-at = {2009-06-22 08:38:18}, priority = {4}, publisher = {ACM}, title = {Y-lib: a user level library to increase the performance of MPI-IO in a lustre file system environment}, url = {http://dx.doi.org/10.1145/1551609.1551617}, year = {2009} }

TY - CONF ID - citeulike:4921049 L3 - citeulike-article-id:4921049 N2 - It is widely known that MPI-IO performs poorly in a Lustre file system environment, although the reasons for such performance are currently not well understood. The research presented in this paper strongly supports our hypothesis that MPI-IO performs poorly in this environment because of the fundamental assumptions upon which most parallel I/O optimizations are based. In particular, it is almost universally believed that parallel I/O performance is optimized when aggregator processes perform large, contiguous I/O operations in parallel. Our research shows that this approach generally provides the worst performance in a Lustre environment, and that the best performance is often obtained when the aggregator processes perform a large number of small, non-contiguous I/O operations. CY - Garching, Germany SN - 978-1-60558-587-1 AD - New York, NY, USA EP - 38 TI - Y-lib: a user level library to increase the performance of MPI-IO in a lustre file system environment PB - ACM T2 - HPDC '09: Proceedings of the 18th ACM international symposium on High performance distributed computing SP - 31 KW - concurrency KW - multicore KW - scalability KW - storage AU - Dickens, Phillip AU - Logan, Jeremy PY - 2009/// UR - http://dx.doi.org/10.1145/1551609.1551617 ER -