Analysis and Evaluation of CAPEX and OPEX in Intra-Data Centre Network Architectures

Abstract

This paper presents a comprehensive analysis of the total cost of ownership (TCO) of intra-data centre networks. A cost model is proposed to evaluate the capital (CAPEX) and operational (OPEX) expenditures that stem from the acquisition and operation of the network infrastructure. The cost model is used to evaluate the TCO of selected network architectures, namely, leaf-spine, fat-tree, hybrid fat-tree, Facebook 4-Post and a new Facebook fabric. The goal is to investigate how the TCO scales for different data centre sizes - given by the number of servers housed in the data centre - and for different server port speeds, i.e. 10GbE, 25GbE, 40GbE and 50GbE. The results show that the most cost-efficient network architecture is leaf-spine. Moreover, for all architectures, the switches are the main cost drivers followed by active optical cables, maintenance and reparation and energy consumption. Regarding the server port speeds, an analysis of economies & diseconomies of scale reveals that 25GbE and 50GbE ports provide an economic migration path to upgrade the network capacity. The reason is that these ports provide lower - or comparable - TCOs per server and per unit of bandwidth than 40 GbE. Besides, a sensitivity analysis shows that the changes in the power consumption caused by the traffic load have small impact on the network TCO.