Burstiness-aware Server Consolidation via Queuing Theory Approach in a Computing Cloud

Burstiness is a common pattern of virtual machines (VMs)´s workload in production data centers, where spikes usually occur aperiodically with low frequency and last shortly. Since virtualization technology enables elastic resource provisioning in a computing cloud, the bursty workloads could be handled effectively through dynamically scaling up/down. However, to cut back energy consumption, VMs are usually highly consolidated with the minimum number of physical machines (PMs) used. In this case, to meet the runtime expanding demands of the resources (spikes), some VMs have to be migrated to other idle PMs, which is costly and causes performance degradation potentially. In this paper, we investigate the elastic resource provisioning problem and propose a novel VM consolidation mechanism with resource reservation which takes burstiness into consideration as well as energy consumption. We model the resource requirement pattern as the popular ON-OFF Markov chain to represent burstiness, based on which a reservation strategy via queuing theory approach is given for each PM. Next we present a complete VM consolidation scheme with resource reservation within reasonable time complexity. The experiment result show that our algorithms improve the consolidation ratio by up to 45% with large spike size and around 30% with normal spike size compared to those provisioning for peak workload, and a better balance of performance and energy consumption is achieved in comparison with other commonly used consolidation algorithms.