Compute and Network Aware VM Scheduling using Reinforcement Learning in Cloud

Cloud computing has been used widely to provide a variety of services which include servers, databases, storage, networking, software, analytics, and intelligence. Scalability is increased and upfront IT infrastructure costs are lowered. In some scenarios, VMs are used extensively for an extended period of time which leads to higher infrastructure costs. Effective virtual machine (VM) management is necessary to make the most out of cloud infrastructures in terms of performance and cost. For scheduling VM effectively we have considered network and compute parameters. Machine learning algorithms are used to predict the best host for the VM placement. We have used reinforcement learning which further uses reward and penalty methods to schedule the VMs, and a methodology is proposed that consists of modules such as Scheduling Module, Host Detection Module, Learning Agent Module, and Consolidation Module. Resource utilization needs to be predicted in order for the system to learn and perform better. The reinforcement-learning-based and ARIMA-based Models are used for better understanding and result analysis. Further, on the basis of three parameters service level agreement violations, energy consumption, and the number of hosts shutdowns, the Reinforcement Learning, and the ARIMA-based Model are compared. The presented Reinforcement Learning Model, which integrates the Enhanced Max-Min algorithm to schedule the VMs and host load prediction, provides the best results by lowering energy usage and service level agreement violations.