Thermal-benchmarking for cloud hosting green data centers

Abstract

Thermal efficient usage of cloud hosting data center servers saves cooling energy and helps establish green cloud data centers. To achieve this goal, the data centers must be stress-tested to avail thermal data related to server utilization. The inherent limitations of cloud computing limit the control of cloud data center owner over the workload execution of cloud services except for the infrastructure and therefore thermal-aware workload modeling or thermal benchmarking of cloud infrastructure can fill this gap. Thermal-benchmarking techniques, through manipulation of server utilization, reveal the thermal profiles and thermal statistics of the servers that can be useful for thermal efficient data center management. This paper presents a generic approach to thermal-benchmarking and profiling of cloud hosting data center servers. We propose workload models to generate customizable thermal benchmarks for stress testing of data center servers. Additionally, we use workload traces from Alibaba cloud to generate thermal statistics to show that the proposed thermal benchmarking approach is applicable to any data center workload trace for any data center server.