student profile: Mr James Phung


Thesis work

Thesis title: Application-Agnostic Power Monitoring and Proportional Computing in Virtualized Environments

Supervisors: Young LEE, Albert ZOMAYA

Thesis abstract:

As distributed networks become more complex and new technologies such as containerization are increasingly adopted, there is an increasing need for versatile solutions that can effectively measure the energy consumption incurred by users of such networks and use this information to schedule user jobs with the goal of maximizing energy efficiency. In this thesis, existing work in power modelling and energy-aware scheduling was reviewed. While much work has been done in these areas, the focus has been on a narrow range of use cases. To address this lack of versatility, three lightweight software-based virtual power meters cWatts+, cWatts++ and RAPL+ were developed. Utilizing simple but powerful application-agnostic power models, they offer similar performances to existing power models but have minimal overheads and are portable for use in a variety of systems. cWatts+ and cWatts++ use a small number of widely available CPU event counters and the Performance Application Performance Interface Library to estimate power usage on a per-application basis; RAPL+ uses the RAPL feature of modern Intel CPUs to perform the same task. Additionally, cWatts++ can easily be used in containerized or virtualized environments. The impacts that using different event counters have on estimation performance were evaluated. Also, the use of CPU core temperature data to improve power estimation performance was examined. cWatts+, cWatts++ and RAPL+ have average absolute errors of less than 5 %. Furthermore, an energy-aware framework PowerSave utilizing RAPL was developed. It supports capping overall CPU power use and the maximum CPU temperature. This framework was evaluated experimentally on 50 concurrent Amazon EC2 bare-metal instances. Results show that PowerSave yields significant energy savings and more proportionate energy use relative to workload level compared to the base case.

Note: This profile is for a student at the University of Sydney. Views presented here are not necessarily those of the University.