Thermal-aware Server Provisioning with Switched MPC for HPC Data Centers

摘要

Abstract: This paper aims at implementing dynamic server provisioning and thermal-aware server assignment methods to particular high-performance-computing data centers which processing a large portion of “small size” jobs. A hysteresis server provisioning algorithm is proposed to reactively provide extra idle servers to process unexpected incoming “small size” jobs. Then, we use the thermal-aware server assignment to allocate active servers to high cooling efficiency areas. To address discrete characteristics of server provisioning, a switched model predictive control approach is developed for thermal-aware active servers allocation, and for obtaining the optimal temperature set-points of the cooling system. Simulation results show that significant energy saving can be achieved.