High-Performance Parallel I/O

Parallel I/O is a necessary component of modern Web and database servers, and of data-intensive applications such as multimedia retrieval, visualization and graphics, scientific simulations, and spatial and geographic databases. Parallel hardware storage systems with multiple disks and high-bandwidth switched interconnect are becoming increasingly available to fill this need. However it is a challenge to schedule and coordinate the I/Os of myriad concurrent devices to meet the resource constraints and timing demands of the applications. The goals of this project are to develop scheduling and resource management algorithms for parallel I/O systems, including issues related to prefetching and caching, on-line scheduling, fair service for multiple users, and deadline-constrained real-time parallel I/O. The techniques are being applied to VBR video retrieval and database applications handling large numbers of concurrent, interacting I/Os.

Participants:

• V. Edwards
• O. Ertug
• B. Hodges
• M. Kallahalla
• P. Varman
• M. Walker

Support:

NSF, High-Performance Parallel I/O Systems, 2001-2004
NSF, Buffer Management for Parallel I/O Systems, 1997-2000
Schlumberger Corporation