Abstract: Cloud computing provides an opportunity to dynamically share the resources among the users through virtualization technology. In this paper, a scheme for load balancing is proposed on the basis of dependency among the tasks. CMS consists of three algorithms including Credit-based scheduling for independent tasks, Migrating Task and Staged Task Migration for dependent tasks. The Credit-based method is used for scheduling the independent tasks considering both user priority and task length. Each task will be assigned a credit based on their task length and its priority. In the actual scheduling of the task, these credits values will be considered. Task Migration algorithm is used to guarantee balancing of loads among the virtual machines. Task migration is done such that the tasks get migrated from heavily loaded machines to comparatively lighter ones. Thus, no rescheduling is required. For dependent tasks, the dependencies between tasks are considered and the technique termed as data shuf?ing is used. In data shuffling, a job is divided into several tasks according to the execution order. The method used here is that the tasks in one stage run independently, while the tasks in different stages must be executed serially. Finally the system is simulated and experiments are conducted to evaluate the proposed methods. This work also concentrates on a simulated study among some common scheduling algorithms in cloud computing on the basis of the response times. The algorithms being compared with the work includes: Random, Random Two Choices (R2C) and On-demand algorithms. The evaluations demonstrate that Credit-based scheduling algorithm significantly reduces the response time.
Keywords: Load Balancing, Virtual Machine, Scheduling, Dependency.