Enabling Research using Hybrid HPC Cloud Computing

Опубликовано 18 июля 2018, 21:03

Hybrid HPC has intrigued the research community since the inception of cloud computing. In this session, Gavin Burris (Wharton School), Amy Apon (Clemson University) and Paul Astell (Communication Research Canada) show how they actually deploy cloud-enabled solutions into production. Wharton School made transparent use of the cloud by seamlessly extending their on-premises (Univa Grid Engine) HPC cluster into Amazon’s EC2 with users virtually unaware as to ‘where’ their HPC jobs actually ran. Clemson researchers achieved a remarkable milestone by deploying a Machine Learning model on more than 1,100,000 vCPUs on Amazon EC2 Spot Instances running in a single AWS region. By extending their resource scheduler SLURM, they broke the record for the largest cluster on the cloud. Communication Research Canada are regularly scaling up multi-thousand core clusters to drive large-scale calculations on radio frequency modeling as well as communications equipment optimizations. These techniques of eliminating the queue and accessing additional resources on-demand are helping researchers via accelerating time to science.