Cloud Federation to Elastically Increase MapReduce Processing Resources

MapReduce is a programming model that allows users the parallel processing of large data sets into a cluster. One of its major implementation is the Apache Hadoop framework that couples both big data storage and processing features. In this paper, we aim to make Hadoop Cloud-like and more resilient adding a further level of parallelization by means of cooperation of federated Clouds. Such an approach allows Cloud providers to elastically scale up/down the system used for parallel job processing. More specifically, we present a system prototype integrating the Hadoop framework and CLEVER, a Message Oriented Middleware supporting federated Cloud environments. In addition, in order to minimize overhead of data transmission among federated Clouds, we considered a shared memory system based on the Amazon S3 Cloud Storage Provider. Experimental results highlight the major factors involved for job deployment in a federated Cloud environment.