Shared Memory Based RDD Data Sharing on Spark

Abstract

Apache Spark is an increasingly popular fast big data analytics engine, which focuses on a large-scale data processing. Currently, Spark's memory management mainly oriented to single application, and do not directly support the typical scenarios of multiple data processing applications. In this paper, we propose an extension of Apache Spark, called Shared Memory Spark (SMSpark). SMSpark introduces shared memory based on RDD data sharing between applications. We conducted experiment in a cluster using two typical applications. Experimental results show that compared to Spark, SMSpark gains better performance.