Dynamic Classification of Repetitive Jobs In Linux For Energy-Aware Scheduling: A Feasibility Study

The workload offered to a typical server consists of many repeated tasks. We present and evaluate a feasibility study that shows how repetitive server workloads can be exploited to enhance the server and CPU energy savings realized by state-of-the-art linux power governors. To minimize dramatic modifications to the web server and the core kernel scheduler we exploit the forensic logging capabilities of the Apache server to collect workload specific information and to schedule requests. We use a daemon to collect the classification statistics and control the dynamic voltage frequency scaling (DVFS) setting of the kernel to batch schedule requests with similar characteristics and thus amortize the energy and performance overhead of making changes to the DVFS settings. Our experimental results show that an energy savings of up to ten percent can be realized for the server on the workloads generated by the SPECweb2005 benchmarks. Keywords-Energy Conservation; Power Savings; Process Management;Web Server; Performance Experiment; Scheduling; Measurement, Power Measurements, Energy Reduction