Dr Paul Coddington1, Mr Sam Morrison2, Mr Sengor Kusturica2, Ms Darcelle Malby3, Mr Andy Botting2, Mr Stephen Crawley3, Mr Dylan McCulloch2, Ms Jo Morris3, Ms Sonia Ramza4, Ms Shubhra Dargar4, Ms Carmel Walsh5
1Australian Research Data Commons, University of Adelaide, Adelaide, Australia, 2Australian Research Data Commons, University of Melbourne, Melbourne, Australia, 3Australian Research Data Commons, University of Queensland, Brisbane, Australia, 4Australian Research Data Commons, Monash University, Melbourne, Australia, 5Australian Research Data Commons, University of New South Wales, Sydney, Australia
There is a growing demand for high-end compute infrastructure in the research sector, particularly for GPUs (e.g. for machine learning) or servers with lots of memory. The ARDC Nectar Research Cloud has previously only supported this infrastructure for dedicated use by ARDC Platforms projects or provisioned by Nectar Node partners for their own researchers. This presentation will describe the implementation of a national infrastructure-as-a-service solution in the Nectar cloud, available to all projects that meet the criteria for a national project allocation.
It was clear that demand would be greater than the capacity we could supply for this infrastructure, so just providing the usual cloud on-demand access would not be the best approach to support many users. Instead, these resources are made available through a reservation system, where users make a reservation to instantiate virtual machine instances with virtualised GPUs or very large memory for a specified period of time. The system is built on top of Blazar, the OpenStack reservation service, with the development of additional novel software to improve the functionality and useability, including enhancements to Blazar, a web interface added to the cloud dashboard, and modifications to our custom allocation system.
We will present an overview of the new GPU and large memory compute infrastructure provided in the Nectar cloud, how it has been virtualised, the choice of standard flavors, the development and functionality of the reservation system, examples of how the service is being used, and feedback from the initial users.
Biography:
Paul Coddington is the Associate Director, Research Cloud and Storage at the Australian Research Data Commons. He has been responsible for the Nectar Research Cloud since 2017 and has more than 30 years experience in the eResearch sector.