Ankündigung einer Lehrveranstaltung im Graduiertenkolleg
Termine
Dienstag, 4. April
2006, Zeit: 10:00-11:30, 12:30-14:00 und 14:15-15:45 Uhr
Mittwoch,
5. April 2006, Zeit: 10:00-11:30, 12:30-14:00 und 14:15-15:45
Uhr
Donnerstag, 6. April 2006, Zeit: 10:00-11:30, 12:30-14:00 und
14:15-15:45 Uhr
Ort
TU Berlin, Einsteinufer
17, 10587 Berlin, Raum E-N 180
Outline
Discrete-event simulation has become the most
commonly used tool for performance evaluation of stochastic dynamic
systems in science and engineering, including such complex systems as
modern multimedia telecommunication networks. This is a result of
significant achievements in electronic and computer technologies that
have led to broad proliferation of powerful computers and computer
networks, and significant achievements in software technology, that
have resulted in simple but very efficient human-computer interfaces.
But no technological innovation can replace the responsibility of
simulators for ensuring that their simulation experiments produce
credible final results.
In this course, we will discuss main problems and solutions of quantitative stochastic discrete-event simulation, i.e. the stochastic simulation in which the emphasis is put on statistical correctness of the final results. Whole spectrum of the problems will be covered: from generators of uniformly distributed pseudo-random numbers, which play the role of original sources of randomness in stochastic simulation, to methods of generation of teletraffic in multimedia networks; from methods of controlling precision of the final results in sequential stochastic simulation on single processors to precision control of the final results in distributed sequential stochastic simulation in multiple parallel time streams.
The latter will be discussed in the context of Multiple Replications in Parallel (MRIP) scenario, which allows to speed up stochastic simulation by launching multiple simulation engines cooperating in production of data for central analysers (one central analyser for each performance measure). An implementation of MRIP in a simulation package Akaroa.2, which automatically launches simulation on an arbitrary number of simulation engines and controls precision of an arbitrary number of analysed performance measures, both in terminating and steady-state simulation, will be also discussed.
The course will be concluded by a survey of open research problems of quantitative stochastic simulation.
Content
Introduction to quantitative stochastic simulation: basic terms and definitions.
Survey of random variables and random processes used in performance modelling of telecommunication networks.
Sources of randomness. Theory and practical solutions of:
generators of independent uniformly distributed pseudo-random numbers for simulation on single and multiple processors;
generators of independent non-uniformly distributed pseudo-random numbers;
generators of correlated sequences of pseudo-random numbers, including strongly correlated (self-similar) sequences.
Sequential quantitative stochastic simulation: its principles and implementation in terminating, steady-state, and non-stationary simulation; measures of precision of the final results. Automation of precision control in terminating and steady-state simulation: survey of methods of analysis of mean values, proportions (probabilities) and percentiles. The initial transient period in steady-state simulation: theory and methods for estimating its length.
Variance reduction techniques and methods of rare events analysis in simulation of telecommunication networks .
Two scenarios of parallel simulation: Single Replication in Parallel (SRIP) and Multiple Replications in Parallel (MRIP); Amdahl’s law of MRIP scenario; MRIP as a Variance Reduction Technique.
Quality of the methods used for precision control of the final results in stochastic simulation: sequential coverage analysis and its results.
Applications of MRIP in Akaroa.2.
Survey of open research problems of quantitative stochastic simulation.