Clustering Of Classroom Computers For Academic Research

Abstract

Academic researchers who are performing computational-heavy research may be at a disadvantage if their available systems are not suitable enough or may take a lot of time to produce a result set. Some academic institutions might not have the necessary resources to address the research needs. In this research, we extend our initial concept of clustering Raspberry Pi systems to investigate the viability and requirements needed to utilise school computers to create a Platform as a Service that can serve the needs for undergraduate researchers. A computer cluster utilising twenty-five school mid-level computers was created with a novel quality of service offering and bench-marked for performance while also creating a platform for students to be able to submit multiple neural network models to train them concurrently by utilising Distributed TensorFlow. It was concluded that low-to-medium end computers could be used to create a computer cluster for research purposes, yet a number of factors need to be taken into consideration.