Sampling Workloads with Dynamic Time Scale to Promote the Energy Efficiency of Datacenters
Cheng Hu, Yi Xin Zhou, Ruoyao Ding · 2020
In modern datacenters (DCs), service resources are allowed to be partially supplied, thus maintaining a good energy efficiency. To match the supplied service resources with the resource demand of workload, the demand is estimated according to the status of workload. Then, in accordance with the estimations, the service resources supplied can be adjusted to match the demand. Generally, some workload factors, which are sampled from workloads, are used to reflect the workload status. But the values of a factor can be very different, when using different Time Scales (TSs) to sample the factor. As a result, the TS used can greatly affect the accuracy of estimation results. In this paper, we propose a Dynamic TS Sampling (DTSS) method to sample workload factors with dynamic TSs. DTSS uses fine-grained TSs for workloads with high variability, and coarse-grained TSs for workloads with moderate variability. Thereby, DTSS can obtain the representative values of factors to reflect workloads' status. Accordingly, based on these values, a good estimation can be made on the resource demand of workloads. Finally, by providing suitable service resources for workloads, DTSS significantly promotes the energy efficiency of DCs.