Dynamic Management of CPU Resources Towards Energy Efficient and Profitable Datacentre Operation
The paper titled "Dynamic Management of CPU Resources Towards Energy Efficient and Profitable Datacentre Operation" was accepted at the 25th Workshop on Job Scheduling Strategies for Parallel Processing (JSSPP 2022). The workshop is held in conjunction with IPDPS 2022.
This work introduces an adaptive Virtual Machine (VM) scheduling and node configuration policy considering runtime information on past CPU usage of the VMs, with VM consolidation, Voltage-Freq Scaling (VFS) and CPU operation at reduced voltage margins. Our approach outperforms two state-of-the-art policies (XM-VFS and LrMmt) that assume a priori
workload knowledge and use offline characterization to derive an optimal static target utilization for the workload at hand.