The IOR (Interleaved Or Random) benchmark program developed at LLNL is intended for use in performance testing of parallel file systems for high performance clusters. The software can be used on any platform which provides an implementation of MPI, the Message Passing Interface, and is in fairly wide use in the HPC community.
IOR provides the capability to test aggregate I/O rates via several typical middleware libraries including MPI collective I/O calls and HDF5 library calls, in addition to POSIX I/O calls. Input arguments allow variance of the overall I/O size, individual transfer size, file access mode (single shared file, one file per client), and whether the data is sequentially or randomly accessed. These and other inputs can be used to mimic the I/O patterns of real HPC applications.
The source code is transitioning from the sourceforge location supported in collaboration with William Loewe during the last few years, to the github site listed below. Future releases will be available there, thanks to the support of Christopher Morrone of LLNL.
Previous versions of IOR up through 2.10.3 will remain available at the old site, for the time being.