Minimizing Energy Consumption of MPI Programs in Realistic Environment

TL;DR

This paper models the problem of minimizing energy consumption in MPI programs using dynamic voltage and frequency scaling, considering hardware constraints, and proposes formulations to estimate optimal frequency schedules.

Contribution

It introduces a mixed integer programming approach to accurately model hardware limitations and estimate minimal energy consumption for MPI programs.

Findings

Two formulations for energy minimization are proposed.

The approach effectively models hardware transition latencies.

Feasibility of solutions is discussed for realistic applications.

Abstract

Dynamic voltage and frequency scaling proves to be an efficient way of reducing energy consumption of servers. Energy savings are typically achieved by setting a well-chosen frequency during some program phases. However, determining suitable program phases and their associated optimal frequencies is a complex problem. Moreover, hardware is constrained by non negligible frequency transition latencies. Thus, various heuristics were proposed to determine and apply frequencies, but evaluating their efficiency remains an issue. In this paper, we translate the energy minimization problem into a mixed integer program that specifically models most current hardware limitations. The problem solution then estimates the minimal energy consumption and the associated frequency schedule. The paper provides two different formulations and a discussion on the feasibility of each of them on realistic applications.