Arithmetic Study about Efficiency in Network Topologies for Data Centers

Abstract

Data centers are getting more and more attention due the rapid increase of IoT deployments, which may result in the implementation of smaller facilities being closer to the end users as well as larger facilities up in the cloud. In this paper, an arithmetic study has been carried out in order to measure a coefficient related to both the average number of hopsamong nodes andthe average number of links among devices for a range of typical network topologies fit for data centers. Such topologies are either tree-like or graph-like designs, where this coefficient provides a balance between performance and simplicity, resulting in lower values in the coefficient accounting for a better compromise between both factors in redundant architectures. The motivation of this contribution is to craft a coefficient that is easy to calculate by applying simple arithmetic operations. This coefficient can be seen as another tool to compare network topologies in data centers that could act as a tie-breaker so as to select a given design when other parameters offer contradictory results.

Study on Data Center Network Topologies for Monitoring Data using Edge/Fog Computing

“…Some typical examples commonly used in switch-centric data center topologies are Fat Tree and Leaf & Spine, whereas some instances typically employed in servercentric data center architectures are BCube, DCell and FiConn [75]. Therefore, those five topologies are going to be studied in order to calculate de average number of links between any given pair of nodes within each topology [76], along with some typical centralization and dispersion statistics related to them, in a way to look into the performance offered by each of those instances [77] .…”