Sophoan Sok, Christian Plewnia, Sansiri Tanachutiwat, Horst Lichter:
Optimization of Compute Costs in Hybrid Clouds with Full Rescheduling
Hybrid clouds enable cost optimizations when handling fluctuating workloads. Assuming that the private cloud is cheaper than the pay-per-use public cloud when it is constantly used, a cost-optimal hybrid operation means that the private cloud handles the base workload and the public cloud the workload peaks. The tasks that cause the workload have to be scheduled such that the private cloud’s usage is maximized. Ideally, the scheduler knows the future tasks and can plan for maximizing the private cloud’s usage. However, obtaining such an accurate prediction of the future is not always possible.
We propose an approach called full rescheduling that allows online schedulers to move already running tasks from the public to the private cloud in order to save costs. We further present an assessment method for studying the economic benefit of full rescheduling and apply it in an industry case study for a data processing application. The study’s results show an economic benefit due to an increased usage of the private cloud instead of the public cloud. The method also provides a cost-optimal number of nodes for the private cloud.
The papers will be published in the IEEE SmartCloud 2020.