Job scheduling in a high performance computing environment

Preparing jobs to run within a high performance cluster environment usually involves at least the understanding of a series of compromises that will affect the time taken to process the work and produce useful results. Software generally is architected by domain specialists who design for a particular hardware environment. Good, well-written software usually incorporates `tweaks' or switches that can be externally invoked to take advantage of the different hardware environments likely to be encountered by the software. With the ever changing landscape of computer hardware, it is not uncommon to have to address the way we work with software in order to maximize the capabilities of the software within a new environment. This paper discusses some of the technical challenges encountered when attempting to use software intended for workstation use within a semi-automatic batch cluster (HPC) environment. The paper chronicles the efforts and solutions deployed working with a team of computational chemists actively engaged on ground-breaking work applied to new drug discovery.