TU Berlin

Internet Network ArchitecturesAnalytical Performance Evaluation

Page Content

to Navigation

Analytical Performance Evaluation

The success of computer and communication systems strongly depends on their performance, typically reflected in the perception of speed. Optimizing system performance, subject to a set of resource and cost constraints, is thus a critical design goal for system engineers. An elegant technique to help in this matter is performance evaluation which can be performed either by measurements, simulation, or using theoretical methods. In particular, analytical performance evaluation has the fundamental merit of rapidly leading to rigorous and unequivocal insight into the behavior of systems which can be accordingly tuned and optimized.

Our own research is concerned with extending the theory of the stochastic network calculus, which is a probabilistic extension of the deterministic network calculus conceived by R. Cruz in the early 1990's. Over the past two decades the calculus has established itself as a versatile alternative methodology to the classical queueing theory for the performance analysis of computer and communication networks. Its prospect is that it can deal with problems that are fundamentally hard for queueing theory, based on the fact that it works with bounds rather than striving for exact solutions. We are in particular concerned with various fundamental research problems related to modelling and analyzing networks with flow transformations, or improving the bounds accuracy using refined inequalities. On the long term, we believe that our research can significantly contribute to establishing the stochastic network calculus as an indispensable mathematical tool for the performance analysis of resource sharing based systems.

Selected Publications

On Superlinear Scaling of Network Delays
Citation key BLC-SSND-11
Author Burchard, Almut and Liebeherr, Jörg and Ciucu, Florin
Pages 1043–1056
Year 2011
ISSN 1063-6692
DOI http://dx.doi.org/10.1109/TNET.2010.2095505
Journal IEEE/ACM Transactions on Networking (ToN)
Volume 19
Number 4
Abstract We investigate scaling properties of end-to-end delays in packet networks for a flow that traverses a sequence of $H$ nodes and that experiences cross traffic at each node. When the traffic flow and the cross traffic do not satisfy independence assumptions, we find that delay bounds scale faster than linearly. More precisely, for exponentially bounded packetized traffic, we show that delays grow with Θ (H log H) in the number of nodes on the network path. This superlinear scaling of delays is qualitatively different from the scaling behavior predicted by a worst-case analysis or by a probabilistic analysis assuming independence of traffic arrivals at network nodes.
Download Bibtex entry

Navigation

Quick Access

Schnellnavigation zur Seite über Nummerneingabe