Elastic Service And Deployment Research For Virtual Machines In An IAAS Cloud

Nowadays, Cloud Computing technology trends to develop rapidly and faces the crucial stage, the development status of it will affect its future deeply. As a kind of infrastructure services IaaS is a basic service mode for Cloud Computing. The paper found that there are still a few of issues in the IaaS Cloud by investigation of Cloud status, such as low service’s efficiency, bad users’ experience and so on. Comparing with current situation of IaaS cloud, the superiority of elastic cloud highlights clearly. Thus, the research for elastic cloud is meaningful; it has scientific significance and research value.In this paper present situation and developing trend of elastic cloud have been analyzed comprehensively through lots of related literatures published in home and domestic and overseas documents. There still a lot of troubles on mechanism of elastic cloud service through analyses, they are the direct factors which leads to many general problems like low service efficiency, big latency etc. Virtual machines are the occupier of resources in an IaaS scenario that provide VMs (virtual machine), accordingly the deployment of them will affect the management of resources under IaaS, on the other word, improving deploying solution of VMs will good for resource management. Researches in this paper paid closed attention to mechanism of elastic cloud, to find a breakthrough to improve the efficiency of elastic ability, forecasting based elastic VMs service and deployment of VMs in the IaaS cloud were exploited. Research works were done revolve around these aspects described as follows:(1)Focused on efficiency of elastic cloud service, a novel algorithm based on ARIMA model for loads predicting was proposed. ARIMA is a mature model for times series analysis, it has its advantages comparing with other models; season index is a series of value which depicts the trait of times series data. The proposed algorithm took advantages of ARIMA and season index and it can predict future loads data of Web services that hosted in the IaaS cloud exactly. The quantity of load from application which has been deployed in the IaaS cloud can be calculated through this algorithm. According to the relationship between load and resources, the quantity of virtual machine includes its configuration should be predicted accurately. As the quantity of virtual machine that the cloud system reserved has been decided, the mechanism of elastic cloud will be achieved based on this algorithm.To verify this algorithm, an useful tool has been applied namely LoadRunner which is the industry standard for application performance testing. A part of log that record the traffic of the2ed world cup website has been used as analyzed data. To test whether the quantity of the reserved virtual machine is reasonable, LoadRunner has been used to simulate real users who send request to this application. The result illustrated that the presented algorithm is useful for the elastic cloud.(2)Virtual machine deployment is a comprehensive problem in the IaaS cloud, it involves resource management, the quality of service, security etc. The related literatures have proved that this is a NP-hard problem. There are a lot of algorithms or models which focused on the NP-hard problem can be used for virtual machine deployment issue. The research found that the big trouble about this issue is to find out rational multiple targets to satisfy the practical requirements of the cloud service and these targets should be transformed information that the algorithm recognized. Elastic cloud framework is the object of this research, the research found the influence which caused by deployment issue cannot be ignored. This paper took a lot of factor into account to implement an appropriate method for the elastic cloud service. The research took advantage of the Genetic algorithm (GA) and proposed a novel deployment solution based improved GA.The research object to reduce the number of used physical computers, improve the utility of resources and satisfy SLA through analysis of practice cloud scenario, meanwhile it should be an adaptive method for initial and dynamic deployment for VMs. GA based algorithm was proposed to deal with the mapping relationship among VMs and servers to solve this multi-objects problem. Experiment results show that this method got the targets we set and it can be used to virtual machine deployment initially or dynamically.Researches in this paper provide a new idea for elastic VMs service and deployment; achievements have an important instructive meaning and scientific value.