Performance Evaluation of Single Board Computer for Hadoop Distributed File System (HDFS)

Abstract

In this paper, we compare the performance of Single Board Computer (SBC) cluster between using local storage and using Network Attached Storage (NAS) as Hadoop Distributed File System (HDFS) data storage. We build a Hadoop cluster with one master node and four slave nodes. The node used is SBC with Micro SD Card as local storage. For benchmarking, we use terasort and testDFSIO programs that available on Hadoop. The results indicate that the performance of SBC for the Hadoop framework increased by using NAS. The processing time needed to run a terasort program is faster with NAS than local storage. The greater the data processed, the greater the percentage of processing time. And the comparison of read speed using both is almost the same but write speed when using NAS is almost twice faster than using local storage. We also use the Performance Analysis Tool (PAT) to measure the performance of SBC. PAT collects CPU performance metrics on SBC when running a terasort program. The results show the CPU utilization of SBC is better to use NAS than local storage.