| Abstract: |
|
Integrating Cloud HPC into an on-premise computation environment could provide a seamless HPC experience from support to actual computation.
Policy-based job execution and data movement present the primary problems for integrating cloud and on-premise HPC systems. Jobs should be routed to the cloud based on the workload type, urgency and funding sources, while other jobs should be routed for local execution. Job execution should dynamically provision required cloud resources and terminate resources upon job completion. For cloud execution, associated data sets should be staged before running, and results may be returned on completion. Data staged may reside on premise or public cloud sources.
We present an open source project based on the ccq metascheduler from CloudyCluster to provide policy based job routing and data staging for on premise HPC resources. We will present a project overview and discuss how it relates to technologies such as CloudyCluster and DellEMC’s 2 TIERS™ |
|
