Research On Application Optimization Of PaaS Platform

Cloud Computing uses virtualization technology to make resources as a pool of services, and provides resources as a service in a scalable on-demand manner. Cloud computing is service-oriented, on-demand, dynamic and distribute, aiming to provide services in a simple way. The task of cloud computing provider is to meet the needs of customer’s Service Level Agreement (SLA), and also to improve the utilization rate of cloud resources, saving energy consumption and operation cost. Nowadays, most research focus resource saving on PaaS(Platform as a Service) layer, seldom consider the application characteristics. In general, scheduling strategy should take multiple SLA indicators into account, such as CPU, IO, memory and network. However, existing research generally only guarantees a certain indicator, most of the scheduling algorithm refer to related fields, such as distributed system or grid. At the same time, web applications are often developed by different developers. They all have very different features and resource consumption patterns. Web applications often face the condition of sudden increasing visiting, so it is more difficult to predict its resource consumption in the near future. How to ensure the multiple SLA indicators and reduce resource is key research of this paper.This approach will be mainly used in small and medium-sized private PaaS platforms. Through monitoring and mining application data on the cloud, this paper proposed a scalable deployment optimization framework based on scoring mechanism. This paper mainly focuses on using application resource consumption feature as one of the references of reducing the use of virtual machine, thus improve resource utilization and reduce energy consumption. By monitoring the application for a certain time, we can get resource consumption patterns of this application. After that, we can use the information to predict future resource consumption details and also study whether it is possible to reduce the amount of virtual machines by migrate some applications. For different prediction algorithms, use sliding window technic to improve accuracy. The use of this framework can be extended to achieve a variety of scheduling algorithm and resource prediction algorithm. Furthermore, the user can choose different strategies of prediction method and migration method, and use history scheme to achieve better performance next time. In the framework, multiple SLA indicators are used in prediction and migration.This paper implements a monitoring and migrating tool based on open source PaaS platform Cloudify. The tool implements four modules of monitoring, system judging, prediction module and migration module by using of Cloudify API. In this tool, two prediction algorithms and migration algorithms are implemented. They are neural network prediction, linear regression, AHP based migration and classification base migrating algorithm. The experimental data set is real web access log. At last, this paper use different experiment to verify the proposed framework. The experiment shows that this framework can effectively use application feature to reduce the consumption of resources.