End-to-End Latency Bottleneck Analysis for Multi-class Traffic in Data Center Networks

It is urgent to monitor and manage end-to-end latency for multi-class network traffic to evaluate latency bottleneck in Data Center Networks (DCNs) to achieve high Quality of Experience (QoE) and network performance. To meet the challenge, a multi-class network traffics should be firstly precisely modeled. Nevertheless, due to their frequent transmission among web servers, application servers and data bases in DCNs, it is hard to precisely predict the transmission path of them, which also makes precisely latency analysis becomes difficult. In this paper, we formulate a precise traffic model for DCNs, aiming to analyzing and positioning the bottleneck occurrence in terms of system latency. Firstly, a new multi-class traffic model is formulated with Network Calculus (NC), and three cases with different positioning precision of the bottleneck (i.e., the whole network, the combination of adjacent nodes, single node) is respectively analyzed. Secondly, a novel bottleneck positioning scheme for those three cases is proposed. Simulation results show our proposed traffic model and the bottleneck positioning scheme can reach a precise analysis for latency bottleneck prediction and management for multi-class traffic in DCNs.