We have it easy, but do we have it right?

Abstract

We show two severe problems with the state of the art in empirical computer system performance evaluation, observer effect and measurement context bias, and we outline the path toward a solution.