Maloney J1, Walters S1, Hankel M1
1Research Computing Centre, Brisbane, Australia
Biography:
Joshua is a graduate of the University of Queensland and holds degrees in physics, mathematics and computer science.
Throughout his career, Joshua has applied the principles of excellence, integrity and inclusivity to the field of technology. He has previously worked with GPU graphics programming, GIS and mapping systems, ocean and atmosphere simulation as well as developing video games for major studios.
He is interested in creating joyful human-computer interactions and exploring the feel and function of new and emerging technology.
Joshua has been with the Research Computing Centre since 2025 and contributes to high-performance computing, clinical data capture and cloud computing services.
Abstract:
Our high-performance computing cluster, Bunya, is heterogeneous (made up of different kinds of nodes). Some of our node types contain coprocessor resources that are in high demand among researchers. Although demand is high, our monitoring has shown that these coprocessors are idle for extended periods of time. We implemented a system that uses our existing scheduling framework and monitoring data to automatically detect job activity in real time. When the usage drops below a predefined threshold for a period of time, we cancel the job. Over an observation period, this method resulted in a reduction of idle time by 2-4% depending on the coprocessor type.
