A Decentralized Approach for Virtual Infrastructure Management in Cloud Datacenters

The ever growing complexity of modern data centers for cloud computing - mainly due to the increasing number of users and their augmenting requests for resources - is pushing the need for new approaches to cloud infrastructure management. In order to face this new complexity challenge, many organizations have been exploring the possibility of providing the cloud infrastructure with an autonomic behavior, i.e., the ability to take decisions about virtual machine (VM) management across the datacenter’s physical nodes without human intervention. While many of these solutions are intrinsically centralized and suffer of scalability and reliability problems, we investigate the possibility to provide the cloud with a decentralized self-organizing behavior. We present a novel migration policy with a twofold goal: saving energy (by putting in sleep mode the underutilized nodes of the datacenter), while keeping the load balanced across the working physical machines. Our migration policy is suitable for a distributed environment, where hosts can exchange status information with each other according to a predefined protocol. We evaluate the performance of the approach by means of an ad hoc built simulator. As we expected, although our distributed implementation cannot perform as good as a centralized management, it can significantly contribute to augment the degree of scalability of a cloud infrastructure.