P. Carns, K. Harms, R. Latham, and R. Ross, "Performance Analysis of Darshan 2.2.3 on the Cray XE6 Platform," Technical Memorandum ANL/MCS-TM-331, October 2012. [pdf]
Darshan is a production-quality I/O characterization tool that captures and summarizes the I/O behavior of parallel applications. It records a variety of information with minimal overhead, including access patterns, number of files accessed, and the amount of time consumed by I/O routines. Darshan's lightweight design makes it suitable for full-time deployment for workload characterization of large HPC systems.
Although Darshan was designed for portability, Darshan 2.2.3 is the first release to feature fully integrated support for the Cray XE6 platform, including support for PGI, Cray, Intel, and GNU compilers as well as both static and dynamic linking. This document presents a brief study of Darshan performance and runtime overhead on the Beagle Cray XE6 system operated by the Computation Institute and the Biological Sciences Division of the University of Chicago and Argonne National Laboratory. We find that Darshan introduces negligible end-to-end overhead for I/O-intensive applications and can store characterization data for over 12 million unique files in less than 5 seconds.