In this paper we explored the questions of if, when and how path switching is beneficial in improving end-to-end performance. For this purpose we set up a wide-area multi-homing testbed with rich path diversity afforded by both multiple service providers as well as overlay paths. Using the delay and loss measurement traces collected on this testbed, we investigated the potential benefits of dynamic end-to-end path switching in terms of both delay and loss performance. Our conclusions are two-fold: 1) While there are significant differences in end-to-end delay performance among various paths, such differences are in general fairly static (i.e., long lasting) and thus can be utilized in a straightforward manner. 2) Whereas, loss rates of different paths vary dynamically, and there is no single "dominant" path that always outperforms others. Hence dynamic path switching among different paths can yield potential performance benefits. Furthermore, we demonstrated that such performance benefits can be effectively exploited by using a simple dynamic path switching strategy.
展开▼