Phase Complexity Surfaces: Characterizing Time-Varying Program Behavior

It is well known that a program execution exhibits time-varying behavior, i.e., a program typically goes through a number of phases during its execution with each phase exhibiting relatively homogeneous behavior within a phase and distinct behavior across phases. In fact, several recent research studies have been exploiting this time-varying behavior for various purposes. This paper proposes phase complexity surfaces to characterize a computer program's phase behavior across various time scales in an intuitive manner. The phase complexity surfaces incorporate metrics that characterize phase behavior in terms of the number of phases, its predictability, the degree of variability within and across phases, and the phase behavior's dependence on the time scale granularity