Sydney
November 6-8, 2017

Event Details


HTC service on underutilized OpenStack cloud resources

Cloud by design is provisioned for peak workloads. So it tends to stay underutilized most of the time. Operating a public cloud is costly. Different approaches are being explored to increase the cloud utilization rate. We explore utilizing these idle cloud resources to run High-Throughput Computing (HTC) jobs. HTC jobs are loosely coupled and doesn’t require high communication among compute nodes. This ‘HTC service’ is being made available in the Massachusetts Open Cloud Marketplace, Giji. The service creates VM on preemptible resources, suspends or resumes VM instances for the HTC jobs based on configurable thresholds and current cloud utilization levels. The Slurm Workload Manager is used to manage the HTC jobs. Slurm user experience via its command line is preserved.

Monday, November 6, 4:20pm-5:00pm
Level: Beginner
None
Rajul is a Graduate Student at Northeastern University, Boston and a Research Assistant at Massachusetts Open Cloud(MOC), Boston. His current work deals with performance monitoring of distributed systems. He is also working on building elastic HPC clusters on OpenStack. FULL PROFILE
Massachusetts Open Cloud, software engineer
Hang is currently working as a software engineer at Massachusetts Open Cloud (MOC). He focuses on developing a simplified graphical user interface for MOC, which could be an "AND" dashboard for users to use besides the horizon dashboard.  FULL PROFILE
Massachusetts Open Cloud
Piyanai Saowarattitada is the Director of Engineering & Infrastructure at the Massachusetts Open Cloud (MOC). Her first experience with OpenStack was back in 2012-13 during Folsom/Grizzly time frame where she led an effort to deploy a Quantum based private cloud for development at (then) HP Cloud services. She joined the MOC leadership team in January 2015 to lead a small team of... FULL PROFILE