PERFORMANCE EVALUATION OF CLUSTER BASED HETEROGENEOUS SYSTEMS USING ADAPTIVE TASK TUNING

Abstract

MapReduce deployment in datacenters and clouds presents a number of challenges in terms of attaining optimal task performance. As a result of server replacement and multi-tenant performance, datacenters and clouds are generally very heterogeneous compared to in-house clusters. As most Mapreduce deployments assume homogeneous clusters, heterogeneity can lead to substantial load imbalances in workflow performance and low utilization of clusters. Notwithstanding the optimisations of task scheduling and load balance, adaptive self-tuning on heterogeneous clusters still is inadequately carried out. The interest in delivering a simple scalable and cost-efficient parallel computing solution for large-scale applications has recently been enormsous with cluster computing platforms. Several adaptive methods have been proposed for the scheduling of adapted and homogenous clusters to increase the performance. However, commodity clusters are naturally adaptive self-tuning task-controlled framework, which is generally upgraded to improve clustering hardware and also introduced new quick machines to boost cluster performance.