Proactive Cloud Management for Highly Heterogeneous Multi-Cloud Infrastructures

Alessandro Pellegrini, Pierangelo Di Sanzo, and Dimiter R. Avresky

Abstract:
Various literature studies demonstrated that the cloud computing paradigm can help to improve availability and performance of applications subject to the problem of software anomalies. Indeed, the cloud resource provisioning model enables users to rapidly access new processing resources, even distributed over different geographical regions, that can be promptly used in the case of, e.g., crashes or hangs of running machines, as well as to balance the load in the case of overloaded machines. Nevertheless, managing a complex geographically-distributed cloud deploy could be a complex and time-consuming task. Autonomic Cloud Manager (ACM) Framework is an autonomic framework for supporting proactive management of applications deployed over multiple cloud regions. It uses machine learning models to predict failures of virtual machines and to proactively redirect the load to healthy machines/cloud regions. In this paper, we study different policies to perform efficient proactive load balancing across cloud regions in order to mitigate the effect of software anomalies. These policies use predictions about the mean time to failure of virtual machines. We consider the case of heterogeneous cloud regions, i.e regions with different amount of resources, and we provide an experimental assessment of these policies in the context of ACM Framework.

BibTeX Entry: