Comp. Psy. 101: The psychology behind high performance computing

Abstract

Building software which can deliver high performance consistently, across a range of different computer clusters, is a challenging exercise for developers as clusters come with specialized architectures and differing queuing policies and costs. Given that optimal code configuration for a particular model on any machine is difficult for developers and end-users alike to predict, we have developed a test which can provide instructions for optimal code configuration, is instantly comprehensible and does not bombard the user with technical details. This test is in the form of a 'personality type' resonant with users' everyday experience of colleagues in the workplace. A given cluster is deemed suitable for either development and or production and small/composite models and or large/complex ones. To help users of our software to choose an efficient configuration of the code, we convert the personality assessment result into a series of optimization instructions based on their cluster's personality type.