Reducing the number of response time SLO violations by a cloud-HPC convergence scheduler

摘要

The migration of jobs from HPC platforms to the cloud has received some interest recently. On the first hand, in such migration scenario, the cloud environment is seen as a virtual hardware extension for the HPC platform. On the other hand, job migration from the cloud to an HPC platform is a much less explored topic. Nevertheless, it may be useful in some cases, in particular when the HPC platform has a low resource usage level and the cloud usage level is high. In such a case, migrating jobs from the cloud to HPC has the potential to avoid evicting cloud jobs due to overbooking. This paper proposes a new scheduling strategy for migrating jobs from the cloud to an HPC platform. Preliminary experimental results show that the proposed strategy reduces the number of response time Service Level Objectives (SLO) violations for cloud jobs and has a low impact on the makespan of a set of HPC jobs.