Our admin Helena, in collaboration with the Rechenzentrum Uni Freiburg, has published “The Chimera and the Cyborg: Hybrid Compute: In vivo HPC, Cloud and Container Implementations”.
High Performance Computing (HPC) systems offer excellent metrics for speed and efficiency when using bare metal hardware, a high speed interconnects, and massively parallel applications. However, this leaves out a significant portion of scientific computational tasks, namely high throughput computing tasks that can be trivially parallelized and scientific workflows that require their own well-defined software environments. Cloud computing provides such management and implementation flexibility at the expense of a tolerable fraction of performance. We show two approaches to make HPC resources available in a dynamically reconfigurable hybrid HPC/Cloud architecture. Both can be achieved with few modifications to existing HPC/Cloud environments. The first approach, from the University of Melbourne, generates a consistent compute node operating system image with variation in the virtual hardware specification. The second approach, from the University of Freiburg, deploys a cloud-client on the HPC compute nodes, so the HPC hardware can run Cloud-Workloads using the scheduling and accounting facilities of the HPC system. Extensive use of these production systems provides evidence of the validity of either approach.