Search Results

1. R. Susitaival, Traffic engineering in the Internet: From traffic characterization to load balancing and peer-to-peer file sharing, Doctoral dissertation, Networking Laboratory, Helsinki University of Technology, 2007 (bib)

   Abstract: Traffic engineering refers to the performance optimization of operational networks. On one hand, traffic offered between origin and destination nodes loads the network and on the other hand, this traffic has to be carried in the network in such a way that performance objectives are fulfilled. In this thesis, we study three different problem areas related to traffic engineering covering traffic characterization, load balancing and peer-to-peer (P2P) file sharing. In the first part of this thesis we characterize the measured traffic on a link in the Finnish backbone network Funet. Traffic on the link is first considered as an aggregate and then split into origin-destination pairs based on the IP addresses of the packets. At a fine level of spatial aggregation we identify four typical OD pair representatives called ``Normal'', ``Bursty'', ``Uniform'' and ``Periodic''. In particular, we are interested in how the so-called moving IID Gaussian model fits together with the aggregate link data and the OD pair representatives. The second part of the thesis considers load balancing in different types of networks. The idea of load balancing is to move traffic from congested links to other parts of the network in a well-controlled way. If the traffic demands are known, the load balancing can be formulated as an optimization problem. However, knowledge of traffic demands is often lacking. For that reason we propose an adaptive and distributed algorithm that gradually balances the load by making small changes in the traffic-splitting ratios on the basis of measured link loads. The application of the adaptive algorithm for both MPLS as well as for OSPF networks is considered. By numerical evaluation we find that the adaptive algorithm converges rapidly almost to the optimum. We also develop optimization algorithms that differentiate traffic classes in terms of mean delay. In this thesis, the differentiation is achieved by the use of both routing and WFQ scheduling. Finally, the load balancing of wirelessmultihop networks is considered by formulating a linear optimization problem for the joint optimization of both routing and scheduling of the network. In the third part of the thesis we consider the population dynamics and performance of novel P2P file-sharing networks. First we study the dynamics of distributing a single chunk by a deterministic fluid model as well as by a more detailed Markov model, which makes evaluation of the lifetime of the system possible. Then we extend the Markov chain model to the case of two chunks and compare the performance of different chunk or peer selection policies in terms of the lifetime as well as the download time of the file. Finally, with a spatio-temporal model we assess how much selecting the nearest peer instead of a random one reduces the usage of the resources of the underlying network.