R. Thakur and W. Gropp, "Parallel I/O," Preprint ANL/MCS-P837-0700, July 2000. [pdf]
Many parallel applications need to access large amounts of data. In such applications, the I/O performance can play a significant role in the overall time to completion. Although I/O is always much slower than computation, it is still possible to achieve good I/O performance in parallel applications by using a combination of sufficient amount of high-speed I/O hardware, appropriate file-system software, appropriate API for I/O, a high-performance implementation of the API, and by using that API the right way. We explain these points in further detail.
Morgan Kaufmann is pleased to present material from a preliminary draft of CRPC Handbook of Parallel Computing, edited by J. Dongarra, I. Foster, G. Fox, K. Kennedy, L. Torczon, and A. White, (c) Copyright 2001 Morgan Kaufmann Publishers. This material may not be used or distributed for any commercial purpose without written permission from the Publisher. Please note that this material is a draft of a forthcoming publication, and as such neither Morgan Kaufmann nor the author can be held liable for changes or alternations in the final edition.