Browse Prior Art Database

A Case Study of Simulation as a Computer System Design Tool Disclosure Number: IPCOM000131246D
Original Publication Date: 1978-Oct-01
Included in the Prior Art Database: 2005-Nov-10
Document File: 9 page(s) / 37K

Publishing Venue

Software Patent Institute

Related People

Gary J. Nutt: AUTHOR [+3]


University of Colorado, Boulder

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 11% of the total text.

Page 1 of 9


This record contains textual material that is copyright ©; 1978 by the Institute of Electrical and Electronics Engineers, Inc. All rights reserved. Contact the IEEE Computer Society (714-821-8380) for copies of the complete work that was the source of this textual material and for all use beyond that as a record from the SPI Database.

A Case Study of Simulation as a Computer System Design Tool

Gary J. Nutt

University of Colorado, Boulder

Simulation can be a powerful design tool, allowing system architects and programmers to optimize both hardware and software before a final design is chosen.

Simulation has been widely used in computer performance studies, including those involving system selection, designing new systems, and tuning the software of existing systems.~-3 Experience shows that simulation can be a particularly effective system design tool -- and, for the design of a parallel processor at the University of Colorado, a necessary one.

Design evaluation tools

Merikallio and Holland have identified several tools that may be used to evaluate the performance of alternative computer system designs:'

(1) mathematical analysis using avereage values,

(2) queuing theory models,

(3) discrete simulation models, and

(4) experimentation with prototypes.

Each of these tools has both strengths and weaknesses. The notion of modeling inherently relies on the elimination of unnecessary detail in constructing a model of a real system. As the amount of detail in a model increases, the model generally tends toward realism and tractability, but its cost also increases.

Models that predict performance based only on average values of the independent variables (the input parameters) are relatively easy to analyze; however, they assume that the variance of those variables is small. If the variance is large, then an average-value model will ignore erratic service and request patterns and underestimate the resources required to maintain performance under peak loads.

Queuing models are more precise, since they allow service and request patterns to be described by probability distributions for random values of the

independent variables; that is, they allow the model to incorporate more detail than average- value models. Unfortunately, as the level of detail increases, queuing models become correspondingly more difficult to solve.

Discrete simulation models also incorporate more detail, since a program will be written to handle each component of the model. The input parameters may be derived from probability distributions (as in queuing models), or they may be provided by trace data. The difficulties with simulation models are that they may be hard to validate and expensive to exercise.

IEEE Computer Society, Oct 01, 1978 Page 1 IEEE Computer Volume 11 Number 10, Pages 31-36

Page 2 of 9

A Case Study of Simulation as a Computer System Design Tool

Prototype developme...