Load Balancing Scheme for the Public Cloud using Reinforcement Learning with Raven Roosting Optimization Policy (RROP)

Efficiently distributing the load among the virtual machines/physical machines by preventing the machines from being overloaded or under loaded improves the overall performance of the computation intensive applications in the cloud-computing platform. The load balancing in a distributed environment like cloud is a challenging activity due to the factors like privacy, security, portability, interoperability, reliability, fault tolerance, virtualization, migration rate, resource utilization and so on. In literature, several static and dynamic load balancing schemes exist for the cloud which has inherent issues in terms of pro-activeness, convergence rate, adaptiveness, and learning capability. By keeping these factors, a novel load-balancing algorithm is proposed using reinforcement learning enabled with raven roosting policy. The proposed load balancer can proactively adapt to the dynamic environment of cloud through reinforcement learning and with the application of raven foraging behavior, the successful task completion rate is found to be high meanwhile the response time and blocking probability of machines are noticed as low.