Energy-Efficient Dynamic Clusters of Servers

In information systems, processes requested by clients are performed on servers in a cluster of multiple servers. Here, not only QoS (quality of service) requirements like response time are satisfied but also the total electric power consumed by servers to perform processes has to be reduced. Furthermore, each process has to be reliably performed in a cluster in presence of server faults. In our approach to reliably performing processes, replicas of each process are redundantly performed on multiple servers. The more number of replicas of a process are performed, the more reliably the process can be performed but the more amount of electric power is consumed by the servers in a cluster. Hence, it is critical to discuss how to reliably and energy-efficiently perform replicas of a process on multiple servers. In this paper, we discuss how to reduce the total electric power consumed by servers in a cluster where each request process is passively replicated on multiple servers. Here, one replica of a process is performed on only one primary server while taking checkpoints in secondary servers. If the primary server is faulty, one of the secondary servers takes over the faulty primary server and a replica of the process is performed from the checkpoint on the new primary server. We evaluate the energy-efficient passive replication (EPR) algorithm of a process in terms of the total power consumption of a server cluster and the average execution time and response time of each process in presence of server faults.