Analysis of performance in PelicanHPC-based Clusters

Abstract

The implementation of high-performance clusters has application in environments of grouping of independent physical nodes and in a virtualized manner through specialized middleware or a dedicated Operating System, with the aim of obtaining greater computational resources from basic equipment. Given both options, the question arises: Which implementation yields better results in terms of performance? Therefore, the purpose of this study is to carry out an analysis of the performance of clusters in two scenarios: ten homogeneous computing devices interconnected to a local network. by braided CAT 6a cable and a switch; and on the other hand, ten virtualized machines in the VirtualBox software, with resources distributed equally such as: RAM memory, processors and disk. The methodology adopted was experimental in five repetitive cycles and working based on the average execution time. The test algorithm is a prime number counter coded in C++ language, generating its output file and subsequent execution at the console level on the terminal. on the PelicanHPC Operating System (live CD variant of the Debian OS). The results are initially uniform, working from 1 to 10 nodes, where the result demonstrates more than 30% improvement in the traditional cluster compared to irregular progress in the virtualized cluster, points that will be the subject of descriptive analysis based on the results obtained.