Assessing Power Needs to Run a Workload with Quality of Service on Green Datacenters

Abstract

Datacenters are an essential part of the internet, but their continuous development requires finding sustainable solutions to limit their impact on climate change. The Datazero2 project aims to design datacenters running solely on local renewable energy. In this paper, we tackle the problem of computing the minimum power demand to process a workload under quality of service constraint in a green datacenter. To solve this problem, we propose a binary search algorithm that requires the computation of machine configurations with maximum computing power. When machines are heterogeneous, we face the problem of choosing the machines and their DVFS state. A MILP (Mixed-Integer Linear Programming), to find the optimal solution, and four heuristics that give satisfactory results in a reasonable time are proposed. The bests reach an average deviation from the optimal solution of 0.03% to 0.65%.