The Redundant Delay Time-Based (RDTB) Algorithm for Fault Tolerant Server Clusters

Server cluster systems are widely used to realize high performance and fault-tolerant information systems. In order to make a server cluster fault-tolerant, multiple replicas of each application process are performed on multiple servers. However, the larger electric energy is consumed in a server cluster due to redundant execution of multiple replicas. The redundant power consumption laxity-based (RPCLB) algorithm is discussed to select multiple servers for energy-efficiently performing replicas of each application process so that the total energy consumption of a server cluster can be reduced. Here, a load balancer has to collect a state of every current replica to estimate the energy consumption of each server. It is difficult to precisely estimate the energy consumption of each server in the real environment since the state of each replica on a server is changed during the estimation. In this paper, we propose a novel redundant delay time based (RDTB) algorithm to select multiple servers for redundantly performing each application process so that the total energy consumption of a server cluster can be reduced. Here, it is not necessary to collect a state of every replica to estimate the energy consumption of each server. Hence, the overhead of a load balancer can be reduced in the RDTB algorithm than the RPCLB algorithm. We evaluate the RDTB algorithm compared with the basic round-robin and RPCLB algorithms. In the evaluation, we show the average total energy consumption of a server cluster in the RDTB algorithm is the same as the RPCLB algorithm and can be maximally reduced to 36% of the RR algorithm. We show the average response time of each process in the RPCLB, RDTB, and RR algorithms is the same. In addition, we show the total number of messages exchanged between a load balancer and servers in the RDTB algorithm is the same as the RR algorithm and can be more reduce than the RPCLB algorithm.