Opportunistic Resource Sharing Based Elastic Resource Allocation in a Data Center

摘要

Resource allocation is the primary issue for multi-tenant cloud data centers. Opportunistic resource sharing is an efficient way to optimize resource utilization for data centers. However, the resource collision makes it challenging. In this paper, we introduce a Markov-chain based model (MST) to characterize the dynamical resource requirements for application, instead of the traditional static virtual network. To deal with the resource collision problem, we introduce a waiting time indicator (WTE) to describe the resource usage, and achieve a tradeoff between utilization and performance. Based on the MST model and WTE indicator, we propose a TRRA algorithm. The basic idea is to select the PM with reasonable WTE value for any extra resource requirement from applications. The experimental results show that our algorithm can realize better resource utilization while guaranteeing acceptable application performance.