An Autonomous Architecture for Managing Vertical Elasticity in the IaaS Cloud Using Memory Over-Subscription

Abstract

Elasticity is one of the essential properties in Cloud Computing that meets changeable needs of customers, and improves resource utilization for providers. In this context, oversubscription is a very powerful technique to increase resource utilization level as much as possible, which leads to a maximization of the profit of cloud providers. In this paper, we propose an autonomous architecture based on the MAPE-K control loop and using memory oversubscription to improve operating performance in a cloud infrastructure. The overload caused by oversubscription is mitigated by the live migration technique of VMs as well as the use of the network memory of the various physical machines of the cluster through the network. This latter technique is usually used as a replacement technique for the swapping disc as it has more performance.