In recent years, real-time applications such as voice and video are increasingly taking up a large share of Internet traffic. However, loss and delay occurred in the underlying network can affect the users' perception negatively. Recent research has shown that applications can avoid end-to-end performance degradation if they bypass problematic networks by dynamically choosing an alternative path to send traffic. Such a dynamic path switching strategy requires the support beyond traditional IP routing and effective utilization of Internet path diversity.
The work in this dissertation was motivated by the desire to improve end-to-end application performance. We present our work from two different perspectives. In the first part, we investigate "how" to exploit Internet path diversity to mitigate end-to-end performance degradation, by proposing several heuristics that construct overlay paths using large peer-to-peer networks. We focus our attention on cases where a large number of choices are available. In other words, we investigate how to exploit the underlying Internet path diversity in an effective and scalable way. Then in the second part, we attempt to understand Internet routing events by studying routing update messages observed from multiple vantage points using statistical methods. Our results show that our proposed scheme may effectively construct alternative paths that serve as the candidate of dynamic path switching while incurring reasonable overhead.
