Building and Supporting User Communities with GPU-Accelerated Computing Services

Investigators:

• Georgios Bekiaris (Astronomy and Space Science – CSIRO)
• Giorgos Vernardos (University of Gronigen)
• Christopher Fluke (Swinburne University of Technology)
• Karl Glazebrook (Swinburne University of Technology)

Proposal Summary:

Access to graphics processing units (GPUs) enables new opportunities for accelerating scientific discovery. Access to a distributed cluster of GPUs – operating as a remote service infrastructure – offers even greater potential for accelerating and managing survey-scale computation and analysis. We have developed and deployed remote GPU-services in two application areas: kinematic model fitting from spectral cube data (GBKFitCloud) and generation and analysis of magnification maps for cosmological microlensing studies (GERLUMPH). In this Project, we aim to improve the usability of these two services by:

• Creating an outward-facing User Interface for GBKFitCloud based on our existing internal-facing application programming interface; and
• Providing robust, secure access to the computing infrastructure, to ensure that researchers can securely and safely perform the processing tasks they require.

Both services are now ready to see their usage and uptake expand, but this cannot occur without improvements to accessibility. The innovative GPU-based approaches we have used can provide significant new research opportunities to the Australian, and international, research communities:

• GBKfitCloud is now ready to work with data from ALMA, and the SAMI, HECTOR, and MaNGA surveys, and will soon be augmented to support the ASKAP WALLABY survey; and
• GERLUMPH has recently been extended to work with a Large Synoptic Survey Telescope light-curve simulator.

The expected outcomes of the Project are enhanced capabilities in kinematic modelling from large-scale optical/IR and radio surveys and a new microlensing methodology for studying quasar structure from synoptic observations. As a software development project, the Project provides a valuable technical test case for provision of remote-service GPU computing.