Design And Implementation Of Service Load Auto-prediction In Cloud Platform

Cloud Computing has become a hot topic, lots of studies and applications basedon Cloud Computing appear like mushrooms after rain, it makes Internet enter the“Cloud Era”. Especially the IaaS service mode, it helps users allocate resourceconveniently. Users deploy their application service on the Cloud Platform, they canapply or return resource according to application service’s load. If we could predictthe service’s load, we can not only give users an early warning, but also help usersallocate or recycle resource timely, make service running properly. Recycling resourcetimely can save cost, it is meaningful to take full advantage of resource. Predictionalso save human cost, and avoid the human mistake and bother, makes CloudPlatform intelligently.This paper implements a service’s load prediction on an Elastic Cloud Platformof IaaS. Using the distributed monitoring system Ganglia to collect load data fromplatform’s VM regularly, then save data into the MongoDB. Prediction program willbe package to a standalone module, it can do predict when predict time up. Predictionreads user’s information, and statistic user’s service. Then prediction reads load data,checks and filter the load data, after that, makes them to be a set of training sample.These samples include input and output sample, and predict input sample. Predictiontrains BP neural network by training samples. We can get predict through importingpredict input sample into the BP. Save predict in MongoDB. In order to show user acurve chart of service’s load, this paper use JSP to show them all, we also providesome suggestion about allocating and recycling resource through analyzing the load.