-
Ks
-
July
5,
2001
Projects
Here you can find short abstracts for current and former projects of the Analytical Modeling Research group of our department. Please follow the links in the abstracts for more detailed information.
PEPSY
In the program system
PEPSY
(Performance Evaluation and Prediction System for Queueing NetworkS)
well known procedures and, in addition, those developed
within the department, were implemented for analytical
performance evaluation based on queueing models. PEPSY has a
user-friendly graphic X-Window interface and is used for
performance evaluation and optimization of computer systems as
well as production systems.
The new Windows version WinPEPSY
has a comfortable graphical and textual input
and output as well as nice experimentation environment.
back to the group page
MOSEL
To evaluate the performance of systems, like computer systems,
networks, production lines and so on, the system
has to be specified. Currently there already exist many different
performance evaluation tools, which all have different input
languages. The basic idea of the new language
MOSEL is, to have
only one language which is easy to use and reflects the
real structure of the system the user wants to describe.
MOSEL will be a common frontend for many performance evaluation
tools. The user has only to become familiar with MOSEL and has
not to learn all the different tool specific languages to use
these tools. After the user has specified its system in MOSEL,
the MOSEL-Compiler will produce the appropiate input
description for the specific tools. By giving options the user
may specify which tools he wants to use. Nevertheless MOSEL
also decides which tools are applicable for the given system
specification. Currently MOSEL can produce input descriptions
for the tools MOSES, which is a Markov analyzer, and the tool
SPNP, which is a Stochastic Petri Net Package. It is planned to
use MOSEL as a common description language for other already
implemented performance evaluation tools, like PEPSY or
SHARPE. The MOSEL-Compiler is implemented in the programming
language C by using the scanner and parser tools lex and yacc.
back to the group page
DiffServ
To improve the Quality of Service (QoS) and to achieve a different
treatment of the different traffic classes in the Internet
the Differentiated Services Architecture (DiffServ) is
introduced (absolute, relative and proportional). The packets
belong to different traffic classes as E-commerce, video, video
conferences, voice, multimedia, ... and are marked in the
IP header. In the stations of the internet the packets are
treated differently depending on the marking and the QoS
requirements for the classes. Important QoS parameters
are the queueing delay and the loss rate.
In the Proportional Differentiated Services Architecture
the mean waiting times (queueing delay) of the different classes
should have a given ratio, the delay ratio, for example two,
three or four.
This can be achieved by a scheduler for the packets at the stations
that uses time dependent priorities. It can be shown by analytical
results that this is correct under heavy load conditions.
An improved scheduler was developed that fulfills these requirements
also under medium load conditions by adapting the parameters
of the time dependent priorities to the actual load of the station.
For only two classes exact values for the parameters can be
calculated. For more than two classes the parameters can be
calculated by optimization for example by using genetic
algorithms. Since the optimization is very time consuming
the parameters cannot be calculated online. They are calculated
in advance and then stored in a look-up table.
back to the group page
Performance Evaluation of Telecommunication Systems
Researcher exchange program with the Lajos Kossuth University of Debrecen, Hungary, for the years 2005 and 2006. This program is funded by the BMBF.
back to the group page
GIANT
With the program system GIANT
(Graphic Interactive Analysis Tool), a tool was developed that
would find the most suitable of several parallelization
alternatives of a program and a given hardware structure with
respect to the total processing time.
back to the group page
PETSI
For UNIX-based
multiprocessor operating systems, a new specifications method
was developed to allow a performance evaluation and comparison
of various parallelization alternatives in UNIX based on the
well known Markov procedures. For some questions regarding
performance evaluation, the stochastic Petri network is
especially suitable, for example, when considering machine
break-down or job processing interdependencies. The program
system PETSI (PETrinet SImulation) allows for careful modelling
of the system (especially in production systems) and, with the
help of simulation, a final determination of the performance
values. Because fault toleration often has to accept
performance reductions over comparable non fault-tolerant
systems, the performance evaluation of such systems is an
important task. This is particularly important when developing
distributed operating or multiprocessor systems. Of
increasingly central interest is the integrated analysis of
performance and dependability. The complexity of the models is
even more of a problem than the pure performance evaluation.
Consequently, approximation techniques and more efficient
solution techniques were developed.
back to the group page
Picasso
The project Picasso is a cooperation with DaimlerChrysler.
back to the group page
Performance Modelling of Computer, Communication and Manufacturing Systems
Researcher exchange program with the Technical University of Budapest, the Lajos Kossuth University of Debrecen, Hungary, the University of Trier, Germany and the University of Hamburg, Germany. This program was funded by the BMBF.
back to the group page
About our Server |
Mail to Webmaster |
Navigational Hints |
Search