Load Balancing in Cloud Computing for Independent and Dependent Tasks with Pre-emption
D. Chitradevi, V. Rhymend Uthariaraj · Transylvanian Review · 2016
The aggregated performance of clouds is highly influenced by the way load balancing and dynamic resource allocations are handled. A way of executing the task in the new Virtual Machine from where it got preempted is needed. Non-preemptive tasks cannot be load balanced during run time from the current state. An efficient way to handle such tasks is the need of the hour. The problem is to optimize the cloud utilization by devising a strategy which handles load balancing and task scheduling effectively. Evaluation shows that centralized load balancing poses the threat of single point of failure. Hence a fully distributed load balancing algorithm is to be presented to cope with the load imbalance problem. This work implements and evaluates the new scheduling and load balancing algorithm by considering the capabilities of each Virtual Machine (VM), current load on Virtual Machines, the task length of each requested job and Task independence/interdependencies. This algorithm will minimize the overload on a VM and subsequently it will minimize the task Migrations also. The performance analysis of all the algorithms is done based on the results of simulation using CloudSim simulation tool. The classes of the CloudSim simulator have been extended to utilize the newly written algorithm. The performance analysis proved that the Preemptive Improved Weighted Round Robin (PIWRR) algorithm is most suitable to the heterogeneous / homogenous jobs with heterogeneous resources (Virtual Machines) than the other Round Robin, Weighted Round Robin and Improved Weighted Round Robin algorithms. Moreover, Preemptive Improved Weighted Round Robin algorithm considers the response time as the main QoS parameter.