A performance and energy optimization mechanism for cooperation-oriented multiple server clusters

To ease the management of data centers, system administrators prefer dedicating one cluster system to serve a certain type of application workload and keeping clusters isolated from one another. However, this approach often results in vulnerable quality of service (QoS), low resource utilization, and incurs high energy costs. In this study, we build a mathematical model to characterize the service process of multi-cluster systems, and compare the isolated versus cooperating usage of multi-cluster alliance based on a thorough analysis of the model. Our study shows that the cooperating approach significantly prevails over the isolated approach for both QoS and energy savings. Consequently, we design a cooperating alliance forming algorithm to automate the cooperation process, and propose a new job scheduling algorithm and resource allocation policies to achieve an efficient and fair cooperation in the cluster alliance. We conduct simulations, and the results prove the accuracy of the model, and show that the proposed algorithm and policies significantly improve both performance and energy savings.

► We propose three metrics to quantify the QoS and energy consumption of a cluster system.
► Based on queuing theory, we model the service process of multi-cluster systems.
► We propose a cooperating algorithm and policy for cluster alliance.
► The cooperating algorithm and policy contribute significantly to QoS and energy savings.