Tuning a Cluster System for High Performance Computing in Engineering

This paper presents results of an effort aimed at performance tuning of a relatively small Beowulf type cluster system used for several years as a workgroup HPC facility at the Institute for Computational Civil Engineering at the Cracow University of Technology, Cracow, Poland. Several directions of the tuning process were considered, like tuning and recompilation of the linux kernel (for e.g. native support for Intel's Hyper Threading (HT) technology), tuning of the TCP/IP protocol stack's parameters, activating the channel bonding (link aggregation) on the GigE interconnect, taking advantage of the new add-ons to the TCP/IP specifications like the Task Offload Engine (TOE), Selective Acknowledgements (SACK), etc. Impact of all the modifications employed were extensively tested with netperf/iperf linux network performance testing tools as well as with the HPL and NPB benchmark suites and will be presented in the paper.