Measuring MPI Latency Variance

Point to point latency and bandwidth measurements are often the first benchmarks run on a new machine or MPI implementation. In this paper we show that the common way of measuring latency hides valuable information about the transmission time characteristics. We introduce a simple benchmark to measure transmission time variance and report on the results from a variety of systems.

[1]  R. Brightwell,et al.  A System Software Architecture for High End Computing , 1997, ACM/IEEE SC 1997 Conference (SC'97).

[2]  Ron Brightwell,et al.  The Portals 3.0 Message Passing Interface Revision 1.0 , 1999 .

[3]  Robert A. van de Geijn,et al.  On Global Combine Operations , 1994, J. Parallel Distributed Comput..

[4]  D.E. Culler,et al.  Effects Of Communication Latency, Overhead, And Bandwidth In A Cluster Architecture , 1997, Conference Proceedings. The 24th Annual International Symposium on Computer Architecture.

[5]  David Scott,et al.  A TeraFLOP supercomputer in 1996: the ASCI TFLOP system , 1996, Proceedings of International Conference on Parallel Processing.