Saturday, March 15

1:30 PM-3:30 PM
Lake Superior

High-Performance and Parallel I/O Systems for Scientific Computing

I/O performance has lagged significantly behind the computation and communication performance of high-performance computers. To eliminate this bottleneck, computer scientists are exploring new techniques and evaluating their effectiveness on high-performance computing testbeds nationwide. One of the most promising research developments is the design of a portable abstract-device interface for parallel I/O, which enables users to run applications portably and efficiently on a variety of high-performance systems. Also under investigation are remote I/O techniques for metacomputing systems and the use of multithreading to improve I/O performance in parallel programs. Central to these efforts is the characterization of I/O-intensive applications requirements, which are then used to guide the development of new programming language features, compiler techniques, runtime libraries, operating system support, parallel file systems, and high-performance networking software.

Organizer: Rick L. Stevens
Argonne National Laboratory

1:30 Experience with Parallel File Systems
John Salmon, California Institute of Technology
2:00 Performance Modeling of a Parallel I/O System: An Application Based Approach
Evgenia Smirni, University of Illinois, Urbana-Champaign
2:30 Portable Implementation of Parallel-I/O APIs
Rajeev Thakur, Argonne National Laboratory
3:00 Where Should Collective I/O be Performed? File System or Runtime System
Alok N. Choudhary, Northwestern University

PP97 Homepage | Program Updates|
Registration | Hotel Information | Transportation | Program Overview | Program-at-a-Glance

MMD, 3/5/97