Multihoming is increasingly being employed by large enterprises and data centers as a mechanism to extract good performance from their provider connections. There is a multitude of commercial route control products for optimizing performance over multiple provider links, that a typical large, multihomed stub network can choose from. However, little is known about the various candidate schemes that such products could employ, their design trade-offs and the potential improvement in performance that they can offer.

In this paper, we evaluate a wide range practical schemes that could go into the design of a route control box and analyze their trade-offs. We implement the various schemes on top of a Linux-based Web proxy and perform a trace-based emulation of their relative performance benefits. Our analysis shows that passive and active monitoring based techniques are equally effective and could improve performance by about 30% when compared to using a single provider. We also show that historical measurement samples are not useful to monitor and predict ISP performance and could, in fact, hurt the performance. We also discuss schemes based on NAT and DNS that multihomed stub networks can employ to control the route of incoming data and direct them over specific provider links.