Big-Data in Climate Change Models — A Novel Approach with Hadoop MapReduce

The goal of this work is to present a software package which is able to process binary climate data through spawning Map-Reduce tasks while introducing minimum computational overhead and without modifying existing application code. The package is formed by the combination of two tools, Pipistrello, a Java utility that allows users to execute Map-Reduce tasks over any kind of binary file, Tina a lightweight Python library that building on top of Pipistrello is able to process scientific dataset, including NetCDF files. We benchmarked the combination of this two tools using a test Apache Hadoop Cluster (4 nodes) and a “relatively” small data set (200 GB), obtaining encouraging results. When using larger clusters and larger storage space, Tina and Pipistrello should be able to scale-up and analyse hundreds of Terabytes of scientific data in a faster, easier and efficient way.