Historically, high performance computing and infrastructure clouds were the two paradigms that did not mix. Cloud hardware architectures coupled with virtualization related overhead precluded HPC communities from realizing the benefits associated with the use of elastic infrastructure. However, uniquely flexible and pluggable architecture of OpenStack cloud platform, makes it possible to overcome these drawbacks. Running OpenStack on top of a low latency, infiniband cluster and incorporating features such as bare metal provisioning, allows one to marry traditional HPC system design concepts with elastic, on-demand infrastructure. In this session, a diverse pool of panelists will discuss architectures, use cases and benefits achieved through incorporating OpenStack to solve various HPC problems. We’ll also examine various ways to measure the performance difference between the bare-metal based machine and the virtualized version. We’ll also examine various ways to measure the performance difference between the bare-metal based machine and the virtualized version.
