Sustainability of big data servers under rapid changes of technology

Abstract

A big data server is a computer system designed to store and process many types of unstructured data arriving at a rapid pace. Such data captured from the Internet and Social Networks are crucial for both developed and developing countries to be able to make informed decisions in time. However, sustainability of big data infrastructures and electronic waste are big issues due to the rapid changes in technology. In this paper we evaluate the performance of big data servers on reusable computers in order to evaluate the scalability and feasibility of constructing big data servers using discarded computers that can be procured as low as $40. In particular, we compare virtualized clusters and bare metal clusters of the low-cost recycled computing nodes for their scalability and feasibility. Virtualized environment is often considered for big data infrastructures due to more efficient management of the clusters despite of the performance overheads. Our study shows that virtualized environment is not scalable for low-cost recycled computing nodes. Our performance evaluation shows that the virtualized cluster is 66% slower than the non-virtualized cluster for read operations. For write operations, the virtualized system is 88% slower than the non-virtualized system.