| Author | Ning Liu, Jing Fu, Christopher D. Carothers, O. Sahni, K.E. Jansen and M.S. Shephard |
|---|---|
| Title | Massively Parallel I/O for Partitioned Solver Systems |
| Year | 2010 |
| Journal | Parallel Processing Letters |
| Volume | 6 |
| Pages | 1-17 |
| Issue | 08 |
| Publisher | World Scientific Publishing Company |
| Abstract | This paper investigates approaches for massively parallel partitioned solver systems. Typically, such systems have synchronized “loops” and will write data in a well defined block I/O format consisting of a header and data portion. Our target use for such a parallel I/O subsystem is checkpoint-restart where writing is by far the most common operation and reading typically only happens during either initialization or during a restart operation because of a system failure. We compare two parallel I/O strategies: a synchronized parallel IO library (syncIO), and a “reduced blocking” strategy(rbIO). Performance tests executed on the Blue Gene/P at Argonne National Laboratory using real CFD solver data from PHASTA (an unstructured grid finite element Navier-Stokes solver) show that the sync IO strategy can achieve a read bandwidth of 47.4GB/sec and 27.5GB/sec write bandwidth using 128K processors. The “reduced-blocking” rbIO strategy achieves an actual writing performance of 17.8GB/sec and the perceived writing performance is 166TB/sec using 128K Blue Gene/P processors. Keywords: massively parallel processing, parallel I/O, synchronized reduced blocking, Blue Gene/P. |
| PDF File |  Download |
| DOI Link | S0129626410000302 |