Usage patterns in multi-tenant data centers

Abstract

Data centers, hosted either on-site or by a third party, have become a dominant computing platform. Here, we focus on the usage patterns of in-production data centers that are hosted by a third party and serve several corporate customers. We characterize the data center workload and concentrate especially on the temporal evolution of utilization of basic resource components. We especially focus on the autonomic aspect of this characterization as it can be used to identify how loads across components change in order to identify conditions that can trigger resource reallocation toward better workload management. To this end, we focus on the resource demands of six distinct corporate customers on two specific data centers, highlight the workload diversity across these customers, and especially focus on how resources are used in time scales that range from minutes to days, weeks, and months. This study fills an important gap in our understanding on how data center resources are used and provides helpful insights for the development of autonomous resource management in multi-tenant data centers.