Exhibitions / Démonstrations
REMO - REsearch Model Optimization Package
Michael Syrjakow (University of Karlsruhe) and Helena Szczerbicka
(University of Bremen)
The software tool REMO (REsearch Model Optimization Package) was primarily
developed to automate the process of model optimization. It comprises a
library of powerful direct optimization algorithms which are applicable to
both parameter optimization of analytical and simulation models. Direct
optimization means that only cost function values are required to guide the
optimization process. Additional analytical information like gradients, etc.
is not used. Beside the optimization algorithms REMO provides an interface to
connect it with conventional modelling tools. Together with a modelling tool
REMO enables the user to implement and analyse the model which has to be
optimized, to specify the optimization problem, and to automatically perform
the model optimization process.
Pythia: A Performance Analyzer of Hierarchical Mass Storage Systems
Odysseas I. Pentakalos (NASA GSFC), Daniel A. Menascé (George
Mason University), and Yelena Yesha (University of Maryland)
Hierarchical mass storage systems are becoming more complex each day and
there are many possible ways of configuring them. An extensible,
object-oriented performance analyzer, called Pythia, was designed and
implemented to allow users to easily determine the most cost-effective
configuration for a mass storage system. The tool incorporates a modeling
wizard component that is capable of automatically building a queueing
network model from a mass storage system representation defined through a
graphical editor.
HiQPN-Tool
Falko Bause, Peter Buchholz, and Peter Kemper (Universität Dortmund)
HiQPN-Tool is a tool to specify and analyze Hierarchically combined
Queueing Petri nets (HQPNs), a class of combined Petri net and
queueing network modeling formalisms.
HiQPN-Tool supports a hierarchical model specification
in combination with qualitative and quantitative (performance) analysis,
which massively profits from the hierarchical specification.
Modelling Distributed Software by Layered Queueing and LQNS
Murray Woodside, Greg Franks, Jerry Rolia (Carleton University)
LQNS (Layered Queueing Network Solver) provides analytic or
simulation solutions of models that combine software resources such
as locks or limited prcess threads, with hardware resources such as
CPUs. The layered framework is a general approach to extended queueing
network modelling of such systems, anbd was developed to help address
performance issues in software design problems that arise in
concurrent and distributed systems.
A Workbench for Gathering Workload Data on Software Components
V. Vetland (Norwegian Telecom) and C. M. Woodside (Carleton University)
The Performance Workbench combines a test environment for
gathering performance data with a browser and visualization tools for
analysing the data. It is intended for analyzing the resource
consumption of a complete program, or of reusable software components
which can be combined into a program.
Artifex by ARTIS
Gian Franco Grinzato (ARTIS)
Artifex(TM) is an integrated family of products for designing and developing
complex event-driven software applications. Artifex allows developers to:
graphically design, model and analyze event-driven systems; build simulators
and prototypes of computer-based systems; create real-time simulators of the
environment which surrounds an application to automate its testing.
Artifex also enables rapid prototyping, including graphical user interfaces
(GUIs), and automatically generates the complete application in C code for
workstation-based environments and embedded systems.
Artifex provides a graphical, object-based language which enables designers
to
describe a system's behavior, functionality and structure. The Artifex
language is based on high-level, object-oriented Petri Nets, to which adds
the
concepts of timing, structure and functionality, and is very effective in
describing complex systems.
Artifex allows significant reductions in systems development time and
enhancements in design quality through modeling, analysis and code generation
capabilities.
DBE: A Tool for Trace Driven Memory Simulation
Vesa Hirvisalo (Helsinki University of Technology)
DBE is an experimental tool designed for trace driven simulation of
processor caches and disk buffers. Trace driven simulation is
flexible and requires no special hardware, but generating traces can
be too slow and the resulting traces too large to handle. To overcome
this problem, DBE uses a compile-time trace compaction, which can
yield significantly smaller traces and faster run times than
with the actual trace.
A Suite of Internet-Accessible Analysis Tools
Aad van Moorsel, Reinhard Klemm, Sampath Rangarajan (Bell Labs)
The Internet provides exciting opportunities for
system analysis tool builders to make available
their software. In a typical scenario, the developer
installs the tool on a web server, and any user
can access and use this tool with a web browser.
There no longer is need for downloading, installing,
and porting; in addition, the tool can benefit from
available special-purpose server hardware. To benefit
from these opportunities within Lucent Technologies,
we are setting up an Internet server that provides
company-wide access to a suite of performance and
reliability modeling tools.
MOSES - MOdeling, Specification and Evaluation System
Gunter Bolch and Stefan Greiner (University of Erlangen), and
Helmut Herold (Siemens AG Erlangen)
In this presentation the Markov analyzer MOSES (MOdeling, Specification and
Evaluation System) and the model description language MOSEL (MOdeling,
Specification and Evaluation Language) are demonstrated using practical
examples such as production lines, operating systems or fault tolerant
systems.
The core of MOSEL consists of constructs suitable for the specification of
the possible states of the system and RULE constructs which describe the
possible transitions. This specification method is very compact and enables
the user also to specify large systems.
The Markov analyzer MOSES automatically creates the Markovian system of
equations from the system specification in MOSEL. Seven different
methods are provided for solving this system.
Predicta, a tool for performance prediction of complex
hardware/software systems
Giancarlo Cervetto, Antonio Serra (Performance Research S.r.l.)
The tools presented, called PREDICTA, represents an highly efficient and
low cost
solution of the problem of evaluating the performances of very complex
hardware
and software systems. In those cases the simulation approach presents many
limitations and drawbacks. As first, the cost required to "code" the model.
In
addition, due to the large number of events to be processed, the usage of a
supercomputer is sometime required.
With PREDICTA, the system under analysis is described as a hierarchy of
modules.
The evaluation is then accomplished by means of the following steps:
a symbolic execution of all the configuration related models produces the
demand
profile for each modeled resource; the congestion analysis, by means of
proprietary
convergency algorithm, computes the application(s) response times and
productivities,
and the coefficient of utilization, arbitration times and number of users
waiting
for service for all system components.
SMART: Simulation and Markovian Analyzer for Reliability and Timing
Gianfranco Ciardo and Andrew S. Miner (College of William and Mary)
SMART is a new tool designed to allow various high-level stochastic
modeling formalisms (such as stochastic Petri nets and queueing
networks) to be described in a uniform environment
and solved using a variety of solution techniques,
including numerical methods and simulation.
As SMART is intended as a research tool, it is written in a modular
way that permits the easy integration of new solution algorithms.
Performance Modeling using DSPNexpress
Christoph Lindemann (Technical University of Berlin)
DSPNexpress is a user-friendly software package for performance analysis with
deterministic and stochastic Petri nets (DSPNs). The development of
DSPNexpress has been motivated by the lack of a software package for an
efficient numerical analysis of DSPNs and the complexity requirements
imposed by evaluating design alternatives for hardware and software
components of multicomputer systems. The main scientific contribution of
DSPNexpress lies in its efficient numerical solution component. The package
DSPNexpress solves complex DSPNs without concurrent deterministic
transitions with four orders of magnitude less CPU time than the previously
known numerical method. As a consequence, DSPNexpress is able to calculate
steady-state solutions of complex DSPNs without concurrent deterministic
transitions with reasonable computational effort on a modern workstation. A
parallel implementation of an entirely new method for numerical stationary
analysis of DSPNs with concurrent deterministic transitions is currently
under development.
SPN2MGM: Tool Support for Matrix-Geometric Stochastic Petri Nets
Boudewijn R. Haverkort, Alexander Ost, Henrik Bohnenkamp (RWTH-Aachen)
We present the tool SPN2MGM that can be used to construct and solve
stochastic Petri nets using matrix-geometric techniques. The tool
automatically recognizes the matrix-geometric structure of the Markov
chain underlying the stochastic Petri net, and solves the Markov
chain with these well-known and efficient techniques. To the best of
our knowledge, SPN2MGM is the only tool that connects a high-level
description technique (SPNs) with the advanced matrix-geometric
computational methods.
The user-interface of SPN2MGM has been developed such that it is easy
to evaluate series of models. Also facilities for tabular (Latex) and
grahical output have been incorporated.
A CAD Tool for the Design of Protocol, Petri Nets, FMS, DSP
Daniel Y. Chao (National ChengChi University)
Current applications of the tool include
(1) Protocol modeling, simulation, analysis, reduction, and syn-
thesis (including error-recovery, extended finite state machine)
(2) DSP performance analysis and processor scheduling (single-
and multi- rate)
(3) same as (1) for flexible manufacturing and animation
(4) act as inference engine for Expert systems
(5) Simulation of communication networks such as batcher-bayan
switching networks (ongonig)
(6) Integration of modeling, simulation, analysis and automatic
source code generation.
(7) Computer-aided education for learning algorithms
iSPN: an Integrated Environment for Modeling Using Stochastic
Petri Nets
C. Hirel, S. Wells, R. Fricks, and K. S. Trivedi (Center for Advanced Computing and Communication)
iSPN is a prototype GUI to SPNP (Stochastic Petri Net Package) developed by
Ciardo et al. The development has used the scripting language Tcl (Tool
Command
Language), developed by John Ousterhout, and extension Tk, a toolkit for X
windows.
The major components of the iSPN interface are a Petri Net editor which allows
graphical input of the stochastic Petri nets and an extensive collection of
vizualisation routines to analyze output results of SPNP and aid for
debugging.
The UltraSAN Modeling Environment
D. Deavours, P. Lee, W. Douglas Obal, and W. Sanders (University of Illinois)
UltraSAN is a software package for model-based performance, dependability
and performability evaluation of hierarchically represented systems. At the
highest level, a system is a composed model, and individual components are
given as stochastic activity networks (SANs), a stochastic extension to
Petri nets. Measures of interest are derived by defining reward structures
at the SAN level. If one chooses to analytically solve the model, automatic
state space lumping, using so-called reduced base model construction
techniques, makes it possible to solve very large models. Besides many
transient and steady-state analytical methods, the UltraSAN includes
terminating and steady-state simulators as well as an importance sampling
tool. UltraSAN has found wide-spread use in industry and academia.
GreatSPN2.0: GRaphical Editor and Analyzer for Timed and Stochastic
PNs
G. Franceschinis, R. Gaeta, M. Ribaudo (University of Torino) and G.
Chiola, A. Rodella (University of Genova)
GreatSPN2.0 comprises
(1) a graphical editor for the design of GSPN
(Generalized Stochastic
Petri Nets), DSPN (Deterministic and Stochastic Petri Nets) and SWN
(Stochastic Well-Formed
Nets) models, with the possibility of generating a PostScript or
Encapsulated PostScript
print file and of performing interactive model animation,
(2) structural
analysis modules
(P and T-semiflows, structural bounds, structural conflict and mutual
exclusion, etc.),
(3) state space analysis modules,
(4) interactive/non interactive
simulation,
(5) Markovian analysis (with state space aggregation for SWN models),
(6) bounds computation.
The tool is available for free for universities and non-profit
organizations,
while it can be purchased from the Dipartimento di Informatica,
Universita' di
Torino, Torino, Italy by other organizations/companies.
EPOCA: Environment for Performance Analysis and Evaluation of
Concurrent Applications
S. Donatelli, G. Franceschinis (Univ. di Torino, Italia),
N. Mazzocca, and S. Russo (Univ. di Napoli, Italia)
The EPOCA system allows to design, develop and verify concurrent applications
running on a network of UNIX workstations, using an integrated set of tools
(compiler and linker, GSPN model generator, makefile generator, language run
time support, monitor, post mortem debugger, GSPN analyzer). The novelty in
EPOCA is the integration of stochastic Petri net models into the programming
environment: a key feature of EPOCA is the automatic construction of a GSPN
model of the program, that can be analyzed to study its qualitative and
quantitative properties.
Although EPOCA has considered a specific CSP-like language, DISC, most of its
features are independent of the details of the programming language used for
the implementation of applications.
N-MAP, an Environment for a Performance Oriented Development
Process of Efficient Parallel Programs
Alois Ferscha and James Johnson (Universitaet Wien)
The integration of performance engineering activities
into the very early phases of the development process of parallel
software is attacked with a methodology and a set of
corresponding tools for performance prototyping.
The N-MAP toolset is a fully integrated development
environment for parallel software, with emphasis on the forecast
of parallel program performance based on early prototypes.
N-MAP provides language support
for the rapid prototyping of programs exploiting data parallelism
and/or parallelism at the task level.
Further features/components of N-MAP are an
automated generation of instrumented source
code for simulation and for execution on real parallel processors,
trace analysis, performance visualization
and execution animation, arbitrarily complex (deterministic and
stochastic) workload specifications, fully automated scenario management,
and a graphical user interface.
DEPEND: A Simulation Environment for testing and validating system
dependability
Z. Kalbarczyk, G. Ries, T. Liu, R.K. Iyer (University of Illinois)
DEPEND is a simulation-based environment that allows to evaluate
fault tolerance attributes of computing and communication systems.
It takes as inputs both VHDL and C++ system description and produces as an
output dependability characteristics including fault coverage, availability
and performance. At the core of DEPEND are simulation engines supported by
a fault injector, a set of fault dictionaries, and component libraries.
The fault injector provides mechanisms to inject faults. The component
libraries contain model building blocks which detail functional descriptions
and characteristics. The fault dictionaries embody possible fault effects of
the given fault types, devices, and circuits.
DEPEND employs a hierarchical modeling and simulation approach that is
intended to allow design evaluation by starting at the device-level physical
constructs, through to the chip-level functional behavior, and going up to the
system-level dependability. With the fault dictionaries (that fault effects
on the higher level in terms of a low level fault model) available at
different abstraction levels, faults can be injected at the lower level, such
as device or gate, and the effects can be evaluated at the higher level,
such as chip or system.
The hierarchical simulation approach allows to obtain realistic fault models
for fault injection experiments and thus, facilitates system dependability
testing and validation under conditions that are likely to happen in the
normal operation.
