deTector: a Topology-aware Monitoring System for Data Center Networks

Abstract

Troubleshooting network performance issues is a challenging task especially in large-scale data center networks. This paper presents deTector, a network monitoring system that is able to detect and localize network failures (manifested mainly by packet losses) accurately in near real time while minimizing the monitoring overhead. deTector achieves this goal by tightly coupling detection and localization and carefully selecting probe paths so that packet losses can be localized only according to end-to-end observations without the help of additional tools (e.g., tracert). In particular, we quantify the desirable properties of the matrix of probe paths, i.e., coverage and identifiability, and leverage an efficient greedy algorithm with a good approximation ratio and fast speed to select probe paths. We also propose a loss localization method according to loss patterns in a data center network. Our algorithm analysis, experimental evaluation on a Fattree testbed and supplementary large-scale simulation validate the scalability, feasibility and effectiveness of deTector.