부산대학교 도서관

Big Data can be defined as large data sets which are being generated from different sources like social media, audios, imaging, logging online websites etc. A need exists to process and analyze this huge amount of data to extract meaningful information. This can be a challenging task. Big data exceeds the processing capability of traditional databases to capture, manage, and process the voluminous amount of data. The use of the MapReduce and Spark frameworks are two common approaches that perform data analytics on Big Data. Both frameworks are open source and capable of cluster computing and fault-tolerance. We perform an experimental evaluation to study the performance differences between MapReduce and Spark. Given the conditions of our implementation, we found significant performance differences. We measured the execution times of the common WordCount program, and the PageRank algorithm, on a single node setup. Another metric of evaluation used is the CPU utilization when processing both batch and iterative jobs.