Node scaling scheduling of real-time tasks in a power-aware datacenter

Power consumption of datacenters, especially of large-scale systems, has a growing concern nowadays due to economic and ecological reasons. In this paper, we propose a power-aware mechanism based on DVS that takes into account energy conservation of homogenous clusters and schedulability of deadline-constrained tasks. Although power-aware real-time scheduling heuristics have been intensively studied in the past decade years, they lack a theoretical guideline. From a theoretical perspective, we prove the necessary and sufficient conditions to test whether real-time tasks can be scheduled over a homogenous cluster, and deduce a lower bound for power reduction while guaranteeing the deadlines not violated. Besides, we develop the most power-efficient speed which can minimize the energy consumption. This speed is constant, only depending on the type of processor, and partitions an effective zone for speed scaling in a power-aware cluster. From a practical perspective, we propose a Node Scaling (NS) algorithm applying with other existing heuristics, which can further reduce the power consumption by extending the number of nodes and slowing the processor speed without missing any deadline constraints. Case studies are carried on Intel PXA270 processors. The simulation results show NS is effective for power reduction on the basis of EDF-FF.