Do More Replicas of Object Data Improve the Performance of Cloud Data Centers?

Abstract

Nowadays, more and more researchers have focused on the performance of cloud data centers. Successful development of cloud data center paradigm necessitates the best QoS for the end users and the Mean Response Time (MRT) of the data requests is one of the most important performance indicators that shall be emphasized on. A cloud data center consists clusters of Rawdata Servers (RDS) that can provide raw data retrieval service. For a single data stored in the data center, there may be several RDS with the target raw data replicas. Hence, when a data request arriving, it has many potential data request paths and the system shall determine the best one for it. In this paper, we aim at answering an interesting question: “Do More Replicas of Object Data Improve the Performance of Cloud Data Centers?”, in order to achieve the minimum MRT of all the requests. The target optimal constrained function has been formulated and two novel load balancing algorithms based on virtual routing method has been proposed, which can achieve near-optimal solutions by theoretical proof. We also found distributing the requests for the same objects among several RDS for load balancing purpose, which is widely used in most data centers, would worsen the system performance. We validate our findings via rigorous simulations with respect to several influencing factors and prove that our proposed strategy is scalable, flexible and efficient for the real-life applications.