Search

A broad spectrum of services running on top of an exponentially growing number of interconnected de¬vices makes network operations more complex than ever. New complex network¬wide behaviors, the vari¬ability of desired objectives incorporating different intents into final decisions together with increasing scalability levels require network infrastructure to be more intelligent, expressive, and robust. Usually, these requirements lead to significant operational complexity and an increased cost of network infras¬tructure. Reducing a manageable network state by better exploiting an expensive network infrastructure without compromising flexibility can overcome new levels of operational complexity and scalability constraints. In the era of Software-Defined Networking, the behavior of network elements is defined by packet processing programs. Due to efficiency constraints, these programs consist of one or multiple packet classifiers. In the first part of the thesis, we propose efficient representations of multifield packet classifiers. In particular, we develop efficient combined representations of multiple packet classifiers, approximate classifiers allowing to trade the classification accuracy for additional memory reduction, and methods constructing ternary representations of range-based packet classifiers. In the second part of the thesis, we show how to use resources of the whole network for traffic monitoring problem that allows to address local resource constraints. Computer Science National Research University Higher School of Economics, Moscow, Russia pub