Official URL: http://www.hamilton.ie/publications/rade_stanojevi...

We begin this thesis by generalizing some results related to a recently proposed positive system model of TCP congestion control algorithms. Then, motivated by a mean ¯eld analysis of the positive system model, a novel, stateless, queue management scheme is designed: Multi-Level Comparisons with index l (MLC(l)). In the limit, MLC(l) enforces max-min fairness in a network of TCP flows. We go further, showing that counting past drops at a congested link provides su±cient information to enforce max-min fairness among long-lived flows and to reduce the flow completion times of short-lived flows. Analytical models are presented, and the accuracy of predictions are validated by packet level ns2 simulations. We then move our attention to e±cient measurement and monitoring techniques. A small active counter architecture is presented that addresses the problem of accurate approximation of statistics counter values at very-high speeds that can be both updated and estimated on a per-packet basis. These algorithms are necessary in the design of router-based flow control algorithms since on-chip Static RAM (SRAM) currently is a scarce resource, and being economical with its usage is an important task. A highly scalable method for heavy-hitter identifcation that uses our small active counters architecture is developed based on heuristic argument. Its performance is compared to several state-of-the-art algorithms and shown to out-perform them. In the last part of the thesis we discuss the delay-utilization tradeoff in the congested Internet links. While several groups of authors have recently analyzed this tradeoff, the lack of realistic assumption in their models and the extreme complexity in estimation of model parameters, reduces their applicability at real Internet links. We propose an adaptive scheme that regulates the available queue space to keep utilization at desired, high, level. As a consequence, in large-number-of-users regimes, sacrifcing 1-2% of bandwidth can result in queueing delays that are an order of magnitude smaller than in the standard BDP-bu®ering case. We go further and introduce an optimization framework for describing the problem of interest and propose an online algorithm for solving it.