A Comprehensive Optimization for Performance, Energy Efficiency and Maintenance in Cloud Datacenters

摘要

Currently, cloud datacenters play a more and more important role in providing all kinds of online information services. However, large-scale datacenters incur a high probability of failures especially for service providers built on cheap commodity hardware. Proper maintenance policy can ensure high performance, dependability and low costs. In this paper, we first construct two stochastic models, separately on request service process and server maintenance process. Specific state aggregation schemes are designed to solve the state explosion problem of huge number of servers in the latter process. Due to server failures, the overall performance of cloud is gradually deteriorating. We then comprehensively analyze and evaluate the performance gains, energy consumptions and maintenance costs from an economic perspective. Thereby, a Markov Decision Process (MDP) problem is formulated to obtain the optimal policy on maintenance and speed scaling of servers. Finally, we solve the MDP problem by policy iterations, and experiments are conducted to show the effectiveness and efficiency of our approach.