Energy Efficiency in Testing and Regression Testing -- A Comparison of DVFS Techniques

Abstract

This paper conducts a pilot study on the energy efficiency in software regression testing. Existing techniques that harness the adjustment of CPU frequencies using Dynamic Voltage and Frequency Scaling can be classified into two categories: general and application-specific. However, existing general techniques ignore execution characteristics and existing application-specific techniques require execution profiling. We propose two non-intrusive algorithms (Case Majority and Case Optimal), which exploit an insight on regression test cases to assure efficiency in modified program versions. We conduct experimentation on three medium-size real-world benchmarks over a cycle-accurate power simulator. The empirical results show that applying our proposed techniques in the context of regression testing can effectively save more energy on one benchmark, and does not suffer from lower performance on the other two benchmarks.