Towards Thermal Aware Workload Scheduling in a Data Center

Abstract

Abstract-High density blade servers are a popular technology for data centers, however, the heat dissipation density of data centers increases exponentially. There is strong evidence to support that high temperatures of such data centers will lead to higher hardware failure rates and thus an increase in maintenance costs. Improperly designed or operated data centers may either suffer from overheated servers and potential system failures, or from overcooled systems, causing extraneous utilities cost. Minimizing the cost of operation (utilities, maintenance, device upgrade and replacement) of data centers is one of the key issues involved with both optimizing computing resources and maximizing business outcome.This paper proposes an analytical model, which describes data center resources with heat transfer properties and workloads with thermal features. Then a thermal aware task scheduling algorithm is presented which aims to reduce power consumption and temperatures in a data center. A simulation study is carried out to evaluate the performance of the algorithm. Simulation results show that our algorithm can significantly reduce temperatures in data centers by introducing endurable decline in performance.

Energy-Aware High Performance Computing: A Taxonomy Study

“…Wang et al develops a thermal aware task scheduling algorithm [53], [55] by predicating resource temperatures based on online task-temperature profiles. In their algorithm, an online task-temperature profile is calculated with the preknowledge of task-temperature profile and RC-thermal model [44].…”

Energy-efficient computing for HPC workloads on heterogeneous manycore chips

“…In this way, DVFS can been used to reduce the cooling energy costs of the data center. There has been a significant amount of work on various strategies for reducing the cooling energy of HPC and non-HPC data centers [8,9,37,45,46,53,54].…”

TIGER: Thermal-aware file assignment in storage clusters

“…The ZBD scheme that uses paoching at where the effect of the heat recirculation is ovserved whereas MinHR manages workload in a way that each pod in a data center generates same amount of heat to minimize heat recirculation [5]. Wang et al proposed a way of calculating heat generated by jobs, which are sorted in descending order of their hotness [10]. All the above strategies focused on computing nodes and used linear power model driven by CPU utilization.…”