Modern research computing needs at academic institutions are evolving. While traditional HPC satisfies most workflows, researchers seek sophisticated, on-demand, and self-service control of compute infrastructure. Furthermore, many also seek policy-compliant safe spaces to compute on sensitive or protected data.
To cater to these users, the Minnesota Supercomputing Institute deployed an OpenStack cloud called Stratus. In contrast to typical clouds, Stratus does not manage internal infrastructure; rather, Stratus complements bare-metal, fully-managed HPC, with a self-service model that can accommodate non-traditional computational and storage needs with HPC-like performance.
This talk describes the lessons learned in launching a platform to support research with specific data-use agreements; and also issues concerning accountability, risk acceptance, and the role of project leadership when a large supercomputing facility deviates from its traditional base of support.
The talk will cover a) some background on MSI and its users; b) identifying research requirements which fit poorly with traditional HPC infrastructure, and how an OpenStack cloud was chosen as a solution; c) Cultural changes for institution and staff; d) Planning and testing through to acquisition and deployment; and e) User enrollment, cost-recovery, and discovery of new use-cases.