Publication

Abstract : Hadoop supports processing and storing large amount of data in distributed computing. Big data technology can be leveraged with GPU to accelerate functionality. Hadoop schedulers plays an important role for performance. Present schedulers either works on fixed capacity or fair sharing among nodes without considering computing capabilities. This work proposes a hybrid approach of capacity and priority schedulers. Proposed scheduler is implemented on GPU enabled Hadoop cluster and compared with capacity and fair schedulers. This proposal analyses the performance of present schedulers of Hadoop like Fair sharing, capacity with the CP (Capacity and priority based) scheduler for GPU based Hadoop cluster. Based on experimental results, a proposed scheduler, proposed scheduler gives better. © 2017 IEEE.