Workload classification in multi-vm cloud environment using deep neural network model

In this competitive world, everyone needs to be prepared for future risks and emergency conditions. In a multi-cloud environment users can easily shift from one cloud to another cloud because of the available data and application transfer technologies. Therefore a strong forecast system is mandatory for such conditions and to stop user migration to other clouds. Virtual Machine (VM) plays an important role in effective resource management and cost reduction in cloud infrastructure. Workload prediction in multi-VM is very useful to handle uncertain situations. In this paper, we propose a promising workload prediction technique that can handle the workload from multiple virtual machines. It has a pre-processing and feature selection engine that handles direct data from these virtual machines and the model is strong enough in classifying data based on historical workloads. This classification enables extra knowledge for the cloud vendor to optimize resource usage. This strategy can be used for producing an alarm whenever there is continuously high utilization of resources in the future. Here, our prediction methodology is experimented with a popular real-world Grid Workload Archive (GWA) dataset and it achieves more than 85% prediction accuracy for CPU, Memory and Disk Utilization.