Evaluation of the Extended Improved Redundant Power Consumption Laxity-Based (EIRPCLB) Algorithm

Application processes have to be not only fault-tolerantly but also energy-efficiently performed in presence of server faults in a cluster of servers. In our previous studies, the extended improved redundant power consumption laxity-based (EIRPCLB) algorithm is proposed where a process is actively replicated on multiple servers. While the cluster can be fault-tolerant, the larger amount of electric power is consumed than non-redundant execution. In order to reduce the total power consumption of servers, the total computation time of the replicas has to be reduced. We discuss an EIRPCLB algorithm for force termination and late restart of replicas. Here, as long as one replica successfully terminates, only a smaller part of every other replica is performed. In this paper, we evaluate the EIRPCLB algorithm in terms of total power consumption and the average response time in homogeneous and heterogeneous clusters. We make clear how the total power consumption of a cluster and response time of each process change according to the redundancy of each process and delay time between servers.