Power consumption management under a low-level performance constraint in the Xen hypervisor

Abstract

Virtualization is the main building block of many architectures and systems from embedded computing to large scale data-centers. Managing efficiently computing resources and their power consumption becomes fundamental to optimize the performance of the workloads running on those systems, however, hardware tools like Intel RAPL can only introduce power caps without considering performance. This paper presents a performance-aware power capping orchestrator for the Xen hypervisor. The tool exploits hybrid power management techniques to minimize power consumption respecting a given SLA, leveraging RAPL and overcoming its limitations. Experimental evaluation shows that the proposed approach guarantees good results for almost all the analyzed benchmarks (e.g. CPU-, memory- and IO-bound).