Energy Efficient Scheduling of HPC-jobs on Virtualized Clusters Using Host and VM Dynamic Configuration

摘要

Energy efficient resource management has become a significant concern in virtualized data centers to reduce operational costs and extend systems' lifetime. The opportunity of reducing energy can be achieved by using Dynamic Voltage Frequency Scaling (DVFS) and hosts consolidation. However, energy management of emerging High Performance Computing (HPC) clouds that host CPU-intensive jobs is more challenging. In this work, we present an optimization solution to assuage the trade-offs between energy and acceptance ratio of jobs. To achieve this, we consider the current multicore processor architecture, which supports DVFS scheme. Furthermore, we tailored an energy model for multicore processor based on the number of active cores, the average running frequency, and memory. A power-aware local VM scheduler is also implemented at the host level. Importantly, we show the importance of including static power in the energy model. Finally, we compared our approach with pure DVFS and DVFS with live migration. The results show that our approach outperforms the other approaches in terms of energy, SLA violation percentage, system utilization, and number of finished jobs. As a future work, we will implement this in a heterogeneous cluster and will consider the cost of turning off and on the hosts.