Solving Queueing Network Models in Cloud Provisioning Contexts

Abstract

In recent years the research community and most of cloud users are trying to propose flexible and general mechanisms to determine how much virtual resources need to be allocated to each tier of the applications executed on cloud infrastructures. The objective of this virtual provisioning is twofold: minimizing resources consumption and meeting the service level agreement (SLA). Most of the current cloud provisioning and scaling solutions are based on analytical models of applications, trying to automate the provisioning decisions making "what-if" response time predictions. Queueing network (QN) models have demonstrated to be a good choice in this kind of contexts. In this work we compare, performing an exhaustive set of experiments on a real cloud architecture with a new provisioning mechanism, exact solutions with approximate solutions estimated from bounding techniques in order to obtain conclusions about the most efficient way of solving these models when making cloud provisioning decisions.