Temporal characterization of demands for data movement on parallel programs

Proposes a metric to characterize program demands for data movement at small time scales. This metric is computed statically over SPMD (single-program, multiple-data) programs for a particular number of processors. We build a model to define time scales at which the execution progress of distinct processors can be consider equal, and we use an operation count to define the temporal distance between operations. We apply the metric to a set of programs that are part of the Splash benchmarks and we discuss the results.