Support Vector Regression based Mapreduce Throttled Load Balancer for Data Centers

Abstract

The process of analyzing big data and other valuable information is a significant process in the cloud. Since big data processing utilizes a large number of resources for completing certain tasks. Therefore, the incoming tasks are allocated with better utilization of resources to minimize the workload across the server in the cloud. The conventional load balancing technique failed to balance the load effectively among data centers and dynamic QoS requirements of big data application. In order to improve the load balancing with maximum throughput and minimum makespan, Support Vector Regression based MapReduce Throttled Load Balancing (SVR-MTLB) technique is introduced. Initially, a large number of cloud user requests (data/file) are sent to the cloud server from different locations. After collecting the cloud user request, the SVR-MTLB technique balances the workload of the virtual machine with the help of support vector regression. The load balancer uses the index table for maintaining the virtual machines. Then, map function performs the regression analysis using optimal hyperplane and provides three resource status of the virtual machine namely overloaded, less loaded and balanced load. After finding the less loaded VM, the load balancer sends the ID of the virtual machine to the data center controller. The controller performs migration of the task from an overloaded VM to a less loaded VM at run time. This in turn assists to minimize the response time. Experimental evaluation is carried out on the factors such as throughput, makespan, migration time and response time with respect to a number of tasks. The experimental results reported that the proposed SVR-MTLB technique obtains high throughput with minimum response time, makespan as well as migration time than the state -of -the -art methods.