Identifying Bottleneck Links Using Distributed End-to-end Available Bandwidth Measurements
Alok Shriram, Jasleen Kaur · 2003
Why discover bottleneck links? The ability to identify overloaded network links has utility for both network operators as well as end-users. For network operators, access to an up-to-date snapshot of overloaded links can help quickly detect outages. Continual monitoring of overloaded links can also help verify the efficacy of load-balancing mechanisms used at the edges of a provider network. For end-users that rely on overlay networks for improving end-to-end transport performance, information about the location of bottleneck links can help select good alternate paths. Furthermore, a collection of bottleneck link snapshots with time can be used to answer fundamental questions such as: Does congestion occur frequently? If so, when and on what kind of links does it occur? In this work, we design and evaluate a tool for estimating bottleneck links⎯links with the least amount of available bandwidth (AB) [Jain02]⎯using distributed end-to-end AB measurements. Why end-to-end measurements? Two factors prevent us from monitoring packet transmission at individual links in order to identify bottleneck links. First, access to routers and links is available only to network operators. Furthermore, end-users may not be allowed access to even any collected data due to protective business policies and privacy issues. Second, deploying a monitoring infrastructure⎯that can keep up with heavy traffic on high-speed links⎯at all links in a network is an expensive undertaking. Thus, even network operators may prefer to rely on a tool that uses edge-to-edge measurements to identify heavily-loaded links [Avaya]. The idea of using end-to-end measurements to discover bottleneck links has also been explored in the design of Pipechar