Optimized Thermal-Aware Workload Distribution Considering Allocation Constraints in Data Centers

Abstract

Power management has been increasingly critical for sustainable data centers. One particular aspect that has a strong impact on the power consumed by a data center is how the workload is distributed among its servers. This distribution can be done integrating thermal models that allow balancing cooling needs with computing needs contributing to reduce overall power consumption. In this paper, we present a workload distribution optimization method for homogeneous server environments that minimizes total heat recirculation. We use a parameter to constrain the total contribution of each node to the recirculated heat and we show that such parameter allows fine-grained control over the number of needed servers and consequently over the balance between IT computing power and cooling power needs. Additionally, we incorporate allocation constraints, representing cases where specific workloads must be allocated to a specific subset of servers only, which for example, result from Service-Level-Agreements with data center customers. These constraints are often found in reality but have seldom been considered in the literature. We carry out simulation experiments using measurement data provided by the Bluesim tool [20]. The results show the effectiveness of the proposed approach in controlling the active servers, thus total power, needed for a given workload while meeting allocation constraints.