Traffic-adaptive routing

Traffic in the Internet is known to change over time and exhibit volatile behaviour. One major challenge in communication networks is the problem of handling these traffic demands that are bursty and hard to predict. Current traffic engineering techniques operate on time scale of several hours, which is too slow to react to quick phenomena such as flash crowds or BGP reroutes. The obvious solution, load sensitive routing, is frowned upon, since routing decisions at short time scales can lead to oscillations. This has prevented load sensitive routing from being deployed since the early experiences in Arpanet, the predecessor of today’s Internet. However, recent theoretical results have shown that a re-routing policy based on game theory provably avoids such oscillation and, in addition, can be shown to converge quickly. In the main part of our work, we describe REPLEX, a distributed dynamic traffic engineering algorithm based on this policy. Exploiting the fact that most underlying routing protocols support multiple equal-cost routes to a destination, it dynamically changes the proportion of traffic that is routed along each path. These traffic proportions are carefully adapted utilising information from periodic measurements and information exchanged between the routers. The required signalling overhead is small. Moreover, it can, under some circumstances, be avoided altogether. We evaluate our algorithm in extensive simulations employing traffic loads that mimic realistic bursty TCP traffic whose characteristics are consistent with self-similarity. The simulations show quick convergence and no oscillations with reasonably chosen parameters. Comparative simulations on realistic network topologies show that REPLEX achieves performance improvements that are as good, if not better than those attained with traditional traffic engineering methods. Since REPLEX as well as many other applications rely on collecting traffic statistics, we furthermore describe an algorithm that enables a router to perform this task in an efficient way. Our Expand-and-Collapse algorithm (EaC) is a heuristic that is particularly aimed at platforms like network processors: Under severe memory constraints, it finds the most popular paths in search trees such as those are used in most packet classification algorithms. EaC allows to efficiently find the largest traffic contributors. In addition, can be applied in a wide number of other contexts as well, due to its genericity. Our theoretical and simulation-based analyses show good performance. A sound understanding for the behaviour of the workload traffic is a vital factor in the construction of load-adaptive dynamic routing and traffic engineering systems. We therefore also describe two methodologies for studying the characteristics of Web traffic, which is still a major part of today’s Internet traffic. Our contributions involve on the one hand a methodology for estimating Web traffic demands on a global scale,