VM performance evaluation with functional models
Jan Sinschek, Andreas Sewe, Mira Mezini · 2009
The performance evaluation of virtual machines is notoriously difficult. Therefore, experimental methodology has recently drawn attention, leading to proposals on how to choose benchmarks, interpret results, and detect measurement bias. But this latter task currently relies on the presence of anomalous measurement results, i.e., on outliers, to raise suspicion. We therefore propose the use of functional performance models to detect bias even when benchmark results might appear unsuspicious. Failure to validate the model indicates either bias or a need to refine the model.