CrowdAdaptor: A Crowd Sourcing Approach toward Adaptive Energy-Efficient Configurations of Virtual Machines Hosting Mobile Applications

Abstract

Applications written by end-user programmers are hardly energy-optimized by these programmers. The end users of such applications thus suffer significant energy issues. In this paper, we propose CrowdAdaptor, a novel approach toward locating energy-efficient configurations to execute the applications hosted in virtual machines on handheld devices. CrowdAdaptor innovatively makes use of the development artifacts (test cases) and the very large installation base of the same application to distribute the test executions and performance data collection of the whole test suites against many different virtual machine configurations among these installation bases. It synthesizes these data, continuously discovers better energy-efficient configurations, and makes them available to all the installations of the same applications. We report a multi-subject case study on the ability of the framework to discover energy-efficient configurations in three power models. The results show that Crowd Adaptor can achieve up to 50% of energy savings based on a conservative linear power model.