Multi-VM Provisioning Mechanism Based On Workload Predicting In Cloud Computing

Cloud computing is a new computing model. It can provide a variety of services to customers via the internet. Infrastructure as a Service (IaaS) is a type of cloud computing service wherein enterprise customers outsource their IT infrastructures. Enterprise customers can construct their IT platform in cloud infrastructure without considering to purchase hardware infrastructures and related system software.With the technology getting more mature, more and more enterprise customers choose to use IaaS. IaaS providers usually provide users with VMs with different specifications, and charge based on the usage time of VM. The elasticity of cloud allows enterprise customers dynamically change VM provisioning scheme based on actual load requirements. And the provisioning scheme should not only meet the workload requirements, but also be cost effective.However, it’s difficult to determine the efficient provision scheme. Currently, there are still two challenges. Firstly, there isn’t efficient workload model to help customers to predict the application’s load. Secondly, the adjustment scheme of the current studies may lead to more rental cost due to neglecting the factor of the granularity of billing time.The issue that we address in this paper is how enterprise customers could make an economical provisioning scheme for computing resources.This work is mainly the following aspects:(1) The workload data of a real website was analyzed, which was used to verify the efficiency of different algorithms. Then, we made a research on Amazon’s EC2platform. And it would provide some important parameters, e.g. VM’s price, hardware configuration etc. for the following experiment. Then traditional algorithms were studied, especially focus on the analysis of the linear prediction algorithm.(2) The existing algorithm is improved from two angles and ARIMA-DEC model is proposed. Firstly, it considers the difference between under-provisioning and over-provisioning in order to guarantee the performance of application. Secondly, it introduce "compensation point" to improve the accuracy of forecasting.(3) Research and analysis of the existing VM provisioning models. Propose a new multi-VM provisioning model based on the existing problem we found. Then design a multi-VM provisioning algorithm based on the model. By analyzing the characteristic of Amazon EC2, we introduce the greedy ideas to improve the efficiency of algorithm in a special scenario. And the algorithm can effectively save rental cost comparing to the existing algorithms through experiment.