Thermal Management Considerations for Geographically Distributed Computing Infrastructures

This paper discusses an approach for optimizing the infrastructure thermal performance related to a geographically distributed computing service. Beginning by modeling the total energy costs associated with cooling a distributed environment, the cooling efficiency of a service is evaluated by superposing the piecewise IT workloads that may be delivered from various locations. We find that the total service-level thermal performance can be distinct from the facility- or infrastructure-level thermal performance, which requires a different global thermal management strategy relative to that of single-site environments. The approach is illustrated for a hypothetical example wherein a service is delivered from three different data centers in geographically diverse locations. Depending on the workload characteristics, the optimal distribution of resources across the data centers varies; but through dynamic resource allocation, it becomes possible to support the same service at increased energy efficiencies.