EAVIRA: Energy-Aware Virtual Infrastructure Reallocation Algorithm

The elastic provisioning of Virtual Infrastructures (VIs) enables a dynamic management of cloud resources (computing and communication) in order to meet the hosted application's requirements. Thus, to perform elasticity requests, providers usually rely on reallocation mechanisms and policies. The concerns regarding the environment and the operational costs indicate energy consumption of the data centers as recurring topic in providers policies. Moreover, energy-aware provisioning is beneficial for tenants also. Recent cost models have introduced an implicit incentive to use computing and networking resources just when need to avoid high rent costs. However, reallocation and elasticity requests can unbalance the Data Center (DC) unnecessarily increasing the number of active servers. In this paper we propose EAVIRA algorithm, which takes into account the proportional sharing of CPU usage of DC servers to calculate individual usage costs to: disable idle equipments, and reallocate VIs. EAVIRA acts on online requests for elasticity configuration and performs an offline load balancing, triggered by the Infrastructure as a Service (IaaS) provider. Our experimental analysis indicates a reduction of energy consumption and an increasing on acceptance ratio of allocation requests.

I. Introduction

The economical and administrative benefits bring out by IaaS clouds to tenants led to the wide dissemination of VIs. With the IaaS model, providers offer idle DC capacity as services, amortizing the operational costs and generating revenue with a reduced marginal cost. Each tenant composes a private and isolated VI in accordance with the hosted application's requirements specifying network and compute parameters. Moreover, IaaS tenants rent VIs for various purposes without concern for maintenance in physical resources [1]. A key enabler for cloud adoption is the elastic provisioning of virtual resources. The variation in the processing load of hosted applications requires mechanisms to expand, reduce and reorganize, in number and configuration, the virtual resources previously allocated [2]. Instead of provisioning resources for the worst case scenario, the elastic management guarantees the configuration required to accommodate the dynamic workload reducing rent costs.