Dynamic Sleep State Schedule for Idle Nodes in Clusters

There are often many idle nodes in a computer cluster due to its workload varies with time. Dynamic sleep mechanisms are helpful to save energy wastage caused by such active idle nodes. By scheduling the sleep states of idle nodes, this paper proposes an approach to balance between the cluster's energy consumption and response times. Idle nodes are classified into different groups according to their sleep states. The numbers of nodes in each group are adjusted by tuning the sleep states of the nodes to match the need of computation resources. An effective metric for power management techniques on idle node is introduced. The significance of our approach on energy efficiency (EE) improvement in computer clusters is demonstrated by experiments.