Path inference in data center networks
Kyriaki Levanti, Vijay Gopalakrishnan, Hyong S. Kim, Seungjoon Lee, Emmanuil Mavrogiorgis, Aman Shakil Shaikh · 2012
Abstract—Data centers host a wide-array of critical applications. However, the diversity in the application requirements results in complex network designs. As a result, managing data centers is becoming increasingly difficult. In this paper, We focus on providing one of the key building blocks of network management: the ability to determine how traffic flows in the network. This information is fundamental to many different network management tasks including troubleshooting, capacity planning, and what-if analysis. Towards that end, we present Chartis, a system which performs per-packet path inference in a data center. Chartis takes as input device configurations and the network’s physical topology and outputs the path of a packet in the network. Specifically, Chartis performs per-hop path inference based on a simplified model of layer-3 routing, layer-2 switching, and the most commonly used routing and forwarding mechanisms. To show its diverse applicabilities, we perform path inference within a campus network and within multiple data centers owned by a major cellular service provider. Using routing information collected from these networks, we validate the correctness of the inferred paths. Our results show that Chartis can quickly and accurately determine paths traversed by packets even in complex data center networks, making it a valuable addition to a network operator’s toolbox. I.