Lamers Contributions to Simulation Speed-Up

Rare Event Simulation and Short-Term Dynamic Simulation for Mobile Network Planning
1. Auflage 2008
ISBN: 978-3-8348-9513-4
Verlag: Vieweg & Teubner
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, 165 Seiten, Web PDF

Reihe: Computer Science (R0)

For the simulation of stochastic systems, those techniques are essential which are able to speed up the simulation, i.e. the equivalent of increasing the accuracy that can be reached within a given run-time, especially with a focus on rare events.

Eugen Lamers explains the principle of simulation speed-up in general, demonstrates the technique RESTART for the simulation of rare events, and addresses its efficiency on distributed systems. Due to the character of RESTART, the efficiency is limited by the high communication overhead. The author introduces the Short-Term Dynamic Simulation concept, developed for the planning of mobile radio networks, which enables fast state space coverage along with the consideration of dynamic aspects. He analyses the speed-up potential on a theoretical basis with analytically solvable Markov models for different utilisations. With realistic models for UMTS network planning he shows that the evaluation of dynamic performance measures benefits from the Short-Term Dynamic Simulation technique.

Zielgruppe

Research

Autoren/Hrsg.

Lamers, Eugen

Inhaltsverzeichnis

Statistical Accuracy.- Simulation Speed-up with RESTART.- Distributed RESTART.- Short-Term Dynamic Simulation Concept.- STD Simulation of UMTS Models.- Conclusions.

Leseproben

Chapter 1 Introduction (S. 1-2)

Communication networks of today must meet high demands regarding the Quality of Service (QoS). Simulative performance evaluation of communication networks is an important means for the design and con.guration of such networks. Other methods of performance evaluation are the measurement of an existing real system and the analytic evaluation of a model describing the target system. Measurement is not possible in the design stage of a system, and analytic methods are usually restricted to very simple models of the target system neglecting some or many possibly important aspects.

In simulation, the complexity of the model and the level of details can be defined flexibly, depending on the aspects the investigation is focused on. Simulations of more complex models, however, are slower than simple models, i. e., they require more run time per simulation time unit. But the question is now, what is meant by slow or fast simulation. It has to be clari.ed what is meant by speed-up, and how it can be de.ned. Assume that results of a reference simulation are available and results from another simulation which has been performed with a certain speed-up technique. The question remains how these techniques can be compared, and some comparison parameter is needed.

This comparison parameter has to be taken from some sorted value space to map the state of the simulation progress to. The progress state the reference simulation has reached when finished, the reference progress state, is mapped to a certain reference value of that value space. The simulation progress state of the other simulation to which the reference simulation is compared also needs to be mapped to that same value space. In fact, for each progress state during the simulation a value is needed. These values are intended to re.ect the statistical accuracy, and on the basis of these values, the simulations can be compared.

If the test simulation reaches the reference progress state in less time than the reference simulation, the test simulation is rated as the faster simulation, or the simulation technique is the faster one. Applying a speed-up technique to a simulation now means, a reference statistical accuracy can be reached in shorter time, or a higher statistical accuracy can be reached within the reference time. The simulation user knows the constraints for conducting simulations. Either the user has a maximum of available time for simulation, or a minimum statistical accuracy is required to manifest a certain statement. In any case, a speed-up technique can either save time or increase statistical accuracy.

The criterion for the statistical accuracy used in this thesis is the relative error for a given range of the value space of the evaluated random variable. The so-called local correlation coef.cient makes a signi.cant contribution to this relative error. Chapter 2 explains the effect of the correlation on the state space coverage and the intrinsic simulation speed-up potential resulting from the systematic reduction of the correlation.

Further, a short introduction on parallelisation is given and the evaluation algorithm Limited Relative Error (LRE) is explained which is of fundamental importance for the investigations of all other chapters. Special speed-up techniques for the simulation of rare events are indispensable. Otherwise, reliable simulation results would not be obtainable within reasonable time. A technique called importance splitting and especially a variant of it called RESTART is introduced in chapter 3. It treats regions of the value space differently in order to reach the relative error limit for each region in a comparable time.

Über Autor(innen)

Dr. Eugen Lamers works on quality assurance in the field of automation.