A Consolidation Strategy Supporting Resources Oversubscription in Cloud Computing

Abstract

Consolidation is an approach that optimizes the utilization of the computing resources. Many datacenter virtualization solutions employ it to reduce the number of physical machines (PMs), and thus save money on hardware, cooling, and electricity. Resources oversubscription, as an effective resource management policy is used to free idle resource in cloud. More virtual machines (VMs) can host in the same PM. However, the limited underlying resource (such as last level cache or I/O bandwidth) provided by PM cannot support much more VMs. The performance of VMs may not be guaranteed due to the underlying resources starvation. Meanwhile, the oversubscription strategy is failure for some VMs when they are migrated to a new PM according to consolidation strategy. In this paper, we present a performance-aware consolidation strategy targeting the oversubscribed cloud. It constructs a performance alarm and the corresponding threshold to control the density of consolidation and maintain the performance isolation among the co-hosting VMs. It also finds the appropriate combination of VMs hosting in the same PM to better support oversubscription. We performed our evaluation on a virtual datacenter simulated by Xen. Our evaluation results show that performance alarm effectively stop the performance isolation destroyed. Furthermore, the policy of VM placement reduces the number of used PMs and also protects maximum oversubscription for each VM.