Parallel I/O
Parallel I/O has drawn increasing attention in the last few years as it has become apparent that file-I/O performance rather than CPU performance may be the limiting factor in current as well as future high-performance parallel systems.
Our contributions to improving parallel I/O performance are:
- Collective buffering
Parallel I/O is a mapping problem from the data layout in node memory to the file layout on disks.
- I/O sensitive scheduling and allocation of jobs
New job scheduling strategies have the potential to minimize network contention and thus improve parallel I/O performance in a multiprogrammed environment.
References