Campus HPC network design and monitoring

The needs of the research communities in research institutes and Higher Education (HE) establishments are demanding evermore powerful computing resources for supporting complex scientific and industrial simulation and modeling, manipulating and storage of large quantities of data [6, 9]. In this paper we present our experience at the University of Huddersfield (UoH), UK in developing the HPC systems infrastructure, removing a technical burden from researchers and enabling quicker and more insightful research outcomes. We have designed and implemented the University of Huddersfield, Queens gate Grid (QGG) campus grid [7]. In the process of building QGG systems and optimising its performance, we have designed and implemented a reliable network system infrastructure. The network topology was re-designed in various stages of system deployment resulting in a reduction of the number of switches, routers and network interconnects. This has led to an improvement in data transmission, a reduction in the possibility of bottlenecks and much reduced data loss [2, 9]. The rapid expansion of our campus grid has led us to question the energy efficiency of our HPC systems. Our initial investigation has targeted the transfer of data and power usage with a view to extending this work to incorporate other metrics, which is the subject of further work.