PEPSY-QNS - January 18th, 1996
PEPSY-QNS
PEPSY-QNS is the Performance Evaluation and
Prediction SYstem for Queueing NetworkS.
Its main purpose is to calculate performance values for queueing networks.
Contents
- Contents
- PEPSY-QNS
- Network types
- Node types
- Analysis methods
- Results
- XPEPSY
- What do you need?
- Getting more information
- OnLineHelp
- Publications
- Contact
The following paragraphs describe which networks and node-types
PEPSY-QNS can handle. In addition to this a short
characterization of the available analysis methods and a list of the
performance measures that are calculated is given.
Queueing Networks
- Network types that PEPSY-QNS can handle:
- open queueing networks
- closed networks
- mixed networks containing open and closed job-classes.
with one or more job-classes. Routing is given by visit ratios or
transition probabilities. A change of a jobs class going through
the network is not possible.
- Node types that can be used:
- BCMP-nodes (M/M/m-FCFS, M/G/1-Processor-Sharing,
M/G/.-Infinite-Server, M/G/1-LCFS)
- ./G/m-FCFS
- nodes with preemptive or nonpreemtpive priority policies
- multiple server nodes with different service-stations
(M/M/m-FCFS-ASYM, M/G/m-FCFS-ASYM)
The service process is defined by its first and second moment -
the service rate and the squared coefficient of variation.
Analysis methods
Until now there are more than 50 different analysis methods implemented and
available to PEPSY-QNS.
- Methods that calculate the performance values using
state-probabilities (numerical analysis),
- Methods that calculate the performance values using a
normalisation constant (Convolution, RECAL - Recursion by
Chain ALgorithm),
- Analysis by decomposition (Kuehn, Chylla, Whitt, Pujolle,
Gelenbe, Summ.., Marie, MEM - Maximum Entropy Method),
- Product form approximation (EPM - Extended Product form
Method, diffusion approximation, ...),
- ABA - Asymtotic Bounds Analysis, BJB - Balanced Job Bounds
analysis,
- Methods based on the MVA - Mean Value Analysis (MVA, SCAT -
Self Correcting Approximation Technique) and
- two different simulation methods.
Calculated performance measures
The table of results that has the same format for all methods contains the
following perfomance values:
- throughput (lambda)
- visit ratio (e)
- average service time (1/mue)
- utilization (rho)
- average response time (mvz)
- average number of jobs (maa)
- average waiting time (mwt)
- average queueing length (mwsl)
The results are shown for each node
- for every job class separately and
- summary over all job classes.
These values are listed for each job class for the overall network:
- throughput (lambda)
- average response time (mvz)
- average number of jobs (maa)
How to work with PEPSY-QNS
PEPSY-QNS has two different user interfaces. The first and
more simple way to work with it only needs the standard UNIX(TM)
environment and ASCII-Terminals. The second and more comfortable way is
to use the X-Windows front end called
XPEPSY.
Working with XPEPSY
Starting
Starting the work with XPEPSY when the queueing network is
available resulting from the modelling phase - as it has to be when using
the "normal" PEPSY.
This is the greeting window of
XPEPSY
Creating the queueing network
The main window of XPEPSY contains the
working area.
Here you have to place your queueing-systems and to define the transition
paths. Using the Graphical
Editor Menu you can select the current edit mode, e.g. create
node or
delete object.
After the layout of the network is done, the parameters of the nodes have
to be specified. This is done in the
Node Data Dialog.
After you specified all parameters that are needed, save the queueing
network model and change to the Analysis Mode to perform the
analysis.
Choosing the analysis method
When switching to the Analysis Mode the
Analysis Menu
appears and the feasible methods are listed in it. The methods shown up are
selected out of the internal database by comparing the networks
characteristics with the possibilities of the algorithms. For any method
that is not able to analyze the network the reason why can be queried.
Performing the analysis
Now it is very simple to perform the analysis. Select the method and simply
click on the "Analyze"-button and the program will start the calculations.
They will be shown up a few seconds later in a
results window. You can continue by
selecting further algorithms to compare their results to the ones you've
got already or go back to the
Graphical Editor to change your model for case study purposes.
The program system is available for any UNIX(TM) system. PEPSY-QNS
has been installed successfully on HPUX, SunOS, IRIX, AIX, Ultrix,
SORIX and some other architectures, as well as on several LINUX distributions.
For the installation of the graphical front-end XPEPSY
you need the X-Window-System (at least revision 4) and the Motif widget-set
(version 1.2 or higher).
On-Line-Help
The On-Line-Help-System coming with PEPSY-QNS can be used during
the work with the basic system or the XPEPSY. The texts are
converted and available here.
Publications
Kirschnick, M.: XPEPSY-Handbuch; Universität
Erlangen-Nürnberg; Institut für Mathematische
Maschinen und Datenverarbeitung IV; Tech. Report; Sep. 1993; German:
(compr. Postscript, ~141kB).
Bolch, G., Kirschnick, M.: PEPSY-QNS - Performance Evaluation and
Prediction SYstem for Queueing NetworkS; Universität
Erlangen-Nürnberg; Institut für Mathematische
Maschinen und Datenverarbeitung IV; Tech. Report TR-I4-92-21; Oct. 1992; German:
(compr. Postscript, ~266kB).
Bolch, G., Kirschnick, M.: PEPSY-QNS - Ein Programmsystem zur
Leistungsanalyse von Warteschlangennetzwerken; In: Walke, B., Spaniol, O.
[Eds.]: Messung, Modellierung und Bewertung von Rechen- und
Kommunikationssystemen; Kurzberichte und Werkzeugvorstellungen zur
7.ITG/GI-Fachtagung Aachen, 21.-23. Sep. 1993; Aachener Beiträge zur Informatik
(Bd. 2), Verlag der Augustinus Buchhandlung, Aachen 1993, pp. 216-220.
Kirschnick, M.: The Performance Evaluation and
Prediction SYstem for Queueing NetworkS - PEPSY-QNS; University of
Erlangen-Nuremberg; Institut für Mathematische
Maschinen und Datenverarbeitung IV; Tech. Report TR-I4-94-18; Jun. 1994;
(compr. Postscript, 60kB).
Contact
For further information contact via mail, email or telephone/fax
Dr. G. Bolch
or
S. Kösters.
Done by Matthias Roessler with changes by Stephan Kösters