Cost optimization in cloud provisioning using Particle Swarm Optimization

摘要

A cloud technology has emerged as a prominent workflow computing infrastructure. The need arises to optimize the allocation of resources to cloud provider's customers. An appropriate number of VMs must be created along with the allocation of supporting resources. Moreover, commercial clouds may have many different purchasing options. Finding optimal provisioning solutions is thus an NP-hard problem. Currently, there are many research works discussing the cloud provisioning cost optimization. However, most of the works mainly concerned with task scheduling. In this paper, we proposed a new framework where number of purchased instance, instance type, purchasing options, and task scheduling are considered within an optimization process. In order to identify a solution in a reasonable amount of time, we studied the use of Particle Swarm Optimization (PSO) technique. The decoding scheme is also designed to convert real values in PSO's particles into an integer representing a solution. The initial results show a promising performance in both the perspectives of the total cost and fitness convergence. We believe that our system will be useful in purchasing options decision. Budget can also be accurately estimated for any specified workflow-based application. We believe that the work will benefit the on-demand provisioning of the virtualized resources as a service in the near future.