Big Data Technologies on Commodity Workstations: A Basic Setup for Apache Impala

Abstract

Big Data technologies brought the idea of parallel processing on cheaper commodity servers. When dealing with huge amount of data, instead of migrating to more performant and costly hardware platforms, or buying resources in cloud, it is more affordable to add a number of cheaper servers as nodes for data processing and/or storage. NoSQL data stores, Hadoop ecosystems, NewSQL platforms have proved viable for Big Data storage and processing. In this paper we were concerned with setting up a platform for big data processing using commodity workstations. Many small and medium sized companies have limited resources and their workstations remain unused for more than 12 hours a day. Here Beowulf Cluster Computing could prove useful. Apache Impala was installed as part of a Hadoop distribution on a 9-node cluster. Three TPC-H database schema were loaded for the scale factors of 1, 2 and 10GB. A series of 100 SQL queries were randomly generated and executed for each scale factor. Results were collected and analyzed for determining if the cluster can provide a decent level of data processing performance.