Performance Evaluation: Model or Problem Driven?
Aug. 24, 2021, 15:45-16:45
Going back as far as the mid 1960’s, a variety of performance evaluation techniques have been proposed and used to help solve design and dimensioning questions for computer systems and networks. In the 1970’s a number of workshops on the topic of computer performance evaluation started to emerge, that over the years evolved into successful conference series that last until today. Performance Evaluation has been a well-acclaimed journal since the beginning of the 1980’s. However, some researchers also claim that the field of performance evaluation has developed too much independently from the development of the systems it intends to evaluate, leading to a fairly isolated sub-discipline that not always addresses the questions that really are important. In the first part of my talk I address this claim, which I consider to be true to a large extend, try to analyse why the field has evolved in this way, and do a number of suggestions for overcoming this isolation. In the second part of my talk, I will touch upon a number a recent modelling efforts I have been involved in myself, in which we have tried to be problem-driven, rather than model-driven.Show bio
Boudewijn Haverkort is an experienced academic in the field of the design and application of all sorts of computer systems, having worked in leadership roles in the Netherlands (at the University of Twente the Embedded Systems Institute, and, currently, Tilburg University) and Germany (at the RWTH Aachen), and in a large number of international (European as well as nationally funded) research and innovation projects. Fields of application he has worked in include wireless and wearable communication systems, internet-based systems, smart industry, smart grids, energy efficient data centers and cyber-security of industrial systems. The applications he prefers to work at relate to the great societal challenges we are facing: sustainability, mobility, and safety & security. He has published some 200 scientific papers in journals and at conferences, and held leadership roles in many international conferences. Numerous MSc students as well as 30+ PhD students graduated under his supervision. He is well-versed in governance and leadership roles in higher education institutions, as well as in public-private partnerships. Boudewijn has been director of the independent open innovation institute on Embedded Systems Engineering (2009-2013; now part of TNO), head of the department of computer science at the University of Twente (2008-2009 and 2017-2018) and serves as chairman of the national research and innovation program on big data and its applications (Commit2Data) since 2016. January 2019, he started as dean for the Tilburg School of Humanities and Digital Sciences. In his (rare) spare time, he finds inspiration in art, music, architecture, literature, mountain walking, and mountain biking.
INRIA and UT Austin
Stochastic Geometry based Performance Analysis of Wireless Networks
Aug. 25, 2021, 15:45-16:45
Stochastic Geometry is commonly used for analyzing spectrum sharing in large wireless networks. In this approach, network elements, such as users and base stations, are represented as point processes in the Euclidean plane, and interference fields as spatial shot-noise processes. The analytical machinery of stochastic geometry and basic formulas of information theory can then be combined to predict important spatial statistics of such networks. The talk will first exemplify this approach by showing how to derive the distribution of the Shannon rates obtained by users in two fundamental models, the Poisson dipole model, which is a mathematical abstraction for a large device to device network, and the Poisson-Voronoi model which is an abstraction for a large cellular network. A few variants of these now classical models will be also discussed, like multi-tier cellular networks, or networks leveraging beam-forming. The talk will then exemplify how to introduce birth-and-death type dynamics in this stochastic geometry framework. This will be illustrated through recent results on the simplest model in this class. In this model, users arrive according to a Poisson rain process on the Euclidean plane and leave with a stochastic intensity proportional to their instantaneous Shannon rate.