OpenStack Summit May 2015 Vancouver

Demand for scalable compute capacity is growing across all areas of research. Applications and the skills to maintain them vary considerably between science teams. Fabrics that can adapt to application requirements simplify deployment scenarios. Service replication offers new avenues to grow capacity in scientific work flows.

Traditional campus high performance computing (HPC) systems have had limited capabilities for adapting cluster environments to application requirements. Adding OpenStack to an HPC architecture provides user access to cloud instances they can easily tune to mirror application requirements. Users can replicate app-specific instances giving them an easy path to scale work flows on larger data sets. Additionally, bare metal provisioning can simplify infrastructure management for HPC operators. Hardware fabrics for production HPC, custom applications, and development environments can be allocated through a common framework, providing migration paths from virtual to physical servers.

The research computing group at the University of Alabama at Birmingham is rebuilding their HPC platform to incorporate OpenStack as a core service. An initial experimental OpenStack fabric demonstrated the utility of rapid service deployment and the flexibility to explore alternative implementations. Aligning our campus HPC infrastructure with public cloud models also promises new opportunities to scale capacity on demand and plan growth. This session will share experiences gained deploying our experimental cloud, designs for the integrated fabric, and challenges faced introducing cloud services into traditional operations.