Parallel computing manages huge amounts of data due to a dramatic increase in computing scale. The parallel file system PVFS version2 (PVFS2) realizes a scalable file system for such huge data on a cluster system. Although several MPI tracing tools can check the behavior of MPI functions, tracing PVFS server activities has not been available. Hence, we have missed chances to optimize MPI applications regarding PVFS server activities although effective usage of limited resources is important even in PVFS servers. An off-line performance analysis tool named PIOviz traces both MPI-I/O calls and associated PVFS server activities to assist optimization for MPI applications. Besides, tracing statistical values of PVFS servers such as CPU usage and PVFS internal statistics assists optimization of MPI applications. In this paper, we demonstrate two performance evaluation tests of the HPIO benchmark, and carry out off-line analysis by using PIOviz. The evaluation shows effectiveness of PIOviz in detecting bottlenecks of MPI-I/O.
IOS Press, Inc.
6751 Tepper Drive
Clifton, VA 20124
Tel.: +1 703 830 6300
Fax: +1 703 830 2300 email@example.com
(Corporate matters and books only) IOS Press c/o Accucoms US, Inc.
For North America Sales and Customer Service
West Point Commons
Lansdale PA 19446
Tel.: +1 866 855 8967
Fax: +1 215 660 5042 firstname.lastname@example.org