Power Saving Mechanism for Multi-cluster Resource Manager with Dynamic Loading Prediction Scheduling Algorithm

The “Green Computing,” is especially important and timely. As computing becomes increasingly pervasive, the energy consumption attributable to computing is climbing, despite the clarion call to action to reduce consumption and reverse greenhouse effects. A high performance-computing cluster is a set of computers gathered together and connected with the intent of merging their processing power towards the same goal. In this paper we developed a multi-cluster resource manager with proposed scheduling algorithm to solve this problem. The multi-cluster resource manager can control and balance loading of various heterogeneous clusters more efficiently. To echo today’s energy saving issues, we also proposed a power saving mechanism. Furthermore, this algorithm also considers several properties of the multi-cluster system, including heterogeneous, dynamic adaptation and the dependent relationship of jobs. And, the power saving mechanism can wake up these idle machines when computing power is requested and power down when the job is done.