MILP formulations for spatio-temporal thermal-aware scheduling in Cloud and HPC datacenters

Abstract

This paper focuses on scheduling problems related to the execution of computational jobs in datacenters with thermal constraints. Mixed integer linear programming (MILP) formulations are proposed that encompass both spatial and temporal aspects of the temperature evolution under a unified model. This model takes into account the dynamics of heat production and dissipation in order to schedule jobs at appropriate times on appropriate machines. The proposed MILP formulations are applicable to both high-performance computing (HPC) and Cloud settings, and can target several objectives including energy and makespan minimization, while incorporating the cooling costs and dynamic voltage and frequency scaling capabilities of servers. The applicability and usefulness of our formulations are demonstrated via several HPC and Cloud case-studies.

Research advances on AI-powered thermal management for data centers

“…Oxley et al [16] ensured that the CRAC unit did not exceed a given power constraint. Pierson et al [17] presumed that the CRAC air temperature was constant. Chaudhry et al [18] used the CRAC and explained the effect of a rise in the inlet temperature of each server.…”

“…T amb represents the external temperature of the server, P is the power of the server, R is the thermal resistance of the server, and C is the specific heat capacity of the server. The model was adopted by Ilager et al [15] to estimate the CPU temperature, and Pierson et al [17] utilized the model and employed an analytical thermal model, which takes into account spatial and temporal temperature behaviors. The authors aim to reduce the power consumption by reducing the heat generated by nodes, which can consequently reduce the DC cooling cost [15,17] .…”

“…They also proposed a thermal-aware and power-aware hybrid power consumption model that synchronously takes into account power consumption from computation, cooling, and migration [13] . Pierson et al [17] incorporated DVFS with their model that makes the formulation of the power consumption no longer a linear function. They also used the constant dynamic power model instead of the DVFS in their experiments [17] .…”

“…Pierson et al [17] incorporated DVFS with their model that makes the formulation of the power consumption no longer a linear function. They also used the constant dynamic power model instead of the DVFS in their experiments [17] .…”

Integrating Pre-Cooling of Data Center operated with Renewable Energies

“…To simulate the thermal behavior of the data center building, we adopted a discrete model used in similar works [18], [30].…”