Data prefetching for scientific workflow based on Hadoop

Abstract

Data-intensive scientific workflow based on Hadoop needs huge data transfer and storage. Aiming at this problem, on the environment of an executing computer cluster which has limited computing resources, this paper adopts the way of data prefetching to hide the overhead caused by data search and transfer and reduce the delays of data access. Prefetching algorithm for data-intensive scientific workflow based on the consideration of available computing resources is proposed. Experimental results indicate that the algorithm consumes less response time and raises the efficiency. © 2012 Springer-Verlag Berlin Heidelberg.