Mrs. Sarah Walters1
1The University of Queensland, St Lucia, Australia
Biography:
Sarah Walters is a Research Computing Systems Engineer at the University of Queensland (UQ) with a strong background in Unix and Linux systems administration and programming. With experience spanning various positions within UQ and elsewhere, Sarah brings a wealth of knowledge and expertise to her role at the Research Computing Centre. She is passionate about working on projects that make a difference, particularly in improving researcher outcomes.
Abstract:
Modern scientific research often requires significant computational, data storage and network resources. For a growing number of workloads, budget constraints prevent purchasing large and often underutilised workstations. Increasingly, no single workstation – or even dedicated server – suffices for the scale of the computation or data set size involved. Consequently, a high-bandwidth network connection to large-scale, highly performant storage is often the only practical way to conduct a workflow, simulation or analysis. With the rapid growth in large AI model usage and very large data sets being produced by scientific instruments, more researchers than ever face these challenges.
High Performance Computers (HPCs) have traditionally solved this problem on the command line with batch schedulers. However, they don't readily enable activities requiring visual feedback and manipulation in real time, and many researchers accustomed to graphical desktop environments find the command line daunting or simply impractical for their workflow.
The University of Queensland implemented Open OnDemand to provide a desktop environment on its HPC. This presentation will cover: 1) The need for such a solution, 2) Deciding on and deploying Open OnDemand, 3) Shifting user profiles for HPC, 4) Wins and benefits achieved, and 5) Implications for future HPC hardware expansions and resource management. The talk will highlight exemplars and specific scientific domains that have benefited significantly from the implementation.