PROJEKTY @

Queues with Long Tails: Extended Entropy Maximisation and a Generalization of the Zipf-Mandelbrot Distribution - Key-note speach by prof. Demetres D. Kouvatsos

Demetres D. Kouvatsos

NetPEn – Networks and Performance Engineering Research
School of Computing, Informatics and Media
University of Bradford
Bradford BD7 1DP West Yorkshire, UK

http://www.inf.brad.ac.uk/staff/index.php?type=p&u=demetres

Slides

The key-note slides in pdf format

Abstract

Many studies of Internet traffic revealed that it is characterised by burstiness, self-similarity (SS) and/or long range dependence (LRD) that can be attributed to the long tails of various distributions with power law behaviour, such as those associated with interevent times and queue lengths. These traffic patterns have adverse impact on network’s operation leading to queues and congestion and, thus, are of significant importance towards the prediction of performance degradation and the capacity planning of the network. Long–tailed distributions have been employed to generate traffic flows in simulation studies, which, however, tend to be rather inflexible and computationally expensive. Thus, there is a need to devise new and cost-effective analytic methodologies for  the credible  assessment  of  the  effects of long tailness in network traffic patterns.

The talk will describe a novel analytic framework, based on the optimisation of    generalised  entropy  measures,  such  as   those  proposed  in   the  fields  of Information Theory, Statistical Physics and Quantification Theory, subject to appropriate mean  value  system  constraints. In this context, a generalisation of the Zipf-Mandelbrot (Z-M) distribution will be devised and interpreted as the steady-state probability distribution of queues with long tails and power law behaviour. Moreover, related analytic algorithms will be devised and typical numerical tests will be presented to assess the credibility of the generalised Z-B queue length distribution and the adverse impact of SS and LRD traffic flows on queue performance.

About the speaker

Prof. Demetres Kouvatsos is the Head of NetPEn - Networks & Performance Engineering Research, University of Bradford, UK and a Visiting Professor at CTIF - Center for TeleInFrastruktur, University of Aalborg, Denmark. He pioneered new and cost-effective analytic methodologies, based on queueing, information and graph theoretic concepts, for the approximate analysis of arbitrary queueing network models (QNMs) and their performance modelling applications into the multiservice networks of diverse technology. Professor Kouvatsos acted as the chair of seven IFIP Working Conferences on the ‘Performance Modelling and Evaluation of ATM & IP Networks’ (ATM & IP 1993-1998, 2000). More recently, under the auspices of the EU Network of Excellence Euro-NGI, he acted as the Chair of four HET-NETs 2003-2006 International Working Conferences on the ‘Performance Modelling and Evaluation of Heterogeneous Networks’ and also directed PERFORM-QNMs - the first EU NoE Euro-NGI PhD Course in ‘Performance Engineering and Queueing Network Models’ (Sept. ’06). Some of his latest scholarly activities include the editing and publication of three research volumes by River Publishers (July ’09) focusing on traffic engineering, performance analysis, mobility management and quality-of-service of heterogeneous networks. Moreover, he is currently editing a performance handbook to be published by Springer, based on tutorial papers concerning the ‘Next generation Internet: Performance Evaluation and Applications’. Prof. Kouvatsos is a Recipient of the IFIP Silver Core Award (1997) and served as a member of the Jury of international experts for the Board of Trustees of the Research Foundation – Flanders (FWO-Vlaanderen), Research Council, Belgium (2004). His professional associations include memberships with the EPSRC College, UK and the IFIP Working Groups IFIP WG6.2 on ‘Network and Inter-network Architectures’ and IFIP WG 6.3 on the ‘Performance of Computer Networks’.

---