Run time virtual machine task migration technique for load balancing in cloud

Abstract

Load Balancing is an important aspect of cloud service centers for optimizing the resource utilization. Consumption of excess power in cloud centers can result in monetary wastage. It is critical that, the resources in the cloud center are utilized optimally; so that, both monetary savings and client satisfaction can be achieved. One of the most popular techniques to achieve load balancing is the Virtual Machine (VM) migration technique; wherein, some of the VMs from overloaded Physical Machines (PMs) are migrated to lightly loaded PMs; however, this technique requires excessive time and monetary cost. Recently, a load balancing technique which migrates VM tasks instead of the actual VM was proposed in the literature. This technique was able to overcome some of the limitations of VM migration technique. Here, the overloaded VM does not accept any new task; however, the new tasks are migrated to lightly loaded VMs. Even though this technique migrates extra tasks to achieve VM load balancing, the already overloaded VMs are not relieved from their existing task burden. If some of the existing and suitable tasks in overloaded VMs are migrated, it could improve efficiency of load balancing. In this work, a new run time VM task migration technique is proposed, which migrates tasks from overloaded VMs. The suitable tasks for migration are selected through a discriminant function, which identifies heavy resource consuming and limited execution progressed tasks for migration. Since, it has been shown in the literature that, optimal task-resource mapping is NPhard, Particle Swarm Optimization (PSO) based solution search technique is proposed. This proposed technique substantially reduces computing load, and achieves good power/energy conservation in the overloaded VMs, when compared with the contemporary VM task migration and VM migration techniques.