StarCube: An On-Demand and Cost-Effective Framework for Cloud Data Center Networks with Performance Guarantee

Abstract

In this paper, we propose a resource management framework called StarCube , which guarantees non-blocking resource allocation and topology-preserving reallocation for fat-tree based multi-tenant cloud data centers. With StarCube , each cloud service is allocated an isolated non-blocking virtual network topology, and the topology provisioned to each service is guaranteed logically unchanged during and after virtual machine reallocation. This resource management problem is formulated and proved to be NP-complete. To achieve high resource efficiency in acceptable time, we propose a cost-effective algorithm with polynomial-time complexity based on StarCube for on-demand resource allocation and reallocation. We demonstrate via extensive simulations that the server resources in StarCube -based cloud data centers can be nearly fully utilized with negligible reallocation cost. The results also show that StarCube supports a large variety of service provisioning feasibly and efficiently for cloud data centers of various scales and with dynamic demands. To the best of our knowledge, StarCube is the first solution to allocating and reallocating cloud services for fat-tree networks with guarantee on non-blocking properties.