Managing a growing high performance infrastructure on a fixed budget

Most universities involved in scientific research utilize some sort of High Performance Computing (HPC) facilities. High Performance Computing (often referred to as High Throughput Computing or Scientific Computing) can be defined as using parallel processing to run advanced applications in an efficient manner. These facilities usually contain a variety of hardware configurations including compute clusters (Beowulf clusters), grids or SMP (Symmetrical Multi Processing) computers. At most institutions the HPC resources utilized by faculty and students are provided in one of three ways: Centralized resources managed by University technology departments, decentralized facilities managed by academic departments, external faculties such as the Pittsburgh Super Computing Center.

The management of HPC facilities can be challenging due to specialized nature of the applications, the specialized support needs and the environmental requirements and cost for housing, acquiring and maintaining the hardware. Furthermore, as research problems continue to increase in complexity, computing demands tend to increase as well. There is a delicate balance that must be maintained between faculty/department needs and available funding. How do you effectively meet the needs of departments with varying HPC needs? How do you meet the demand for HPC resources with minimal financial resources? How do you restructure user support to meet the increasing demand?