Arranged by:
Peter Fritzson
and
Christoph Kessler
Programming Environment Laboratory
Department of Computer and Information Science (IDA)
Linköpings universitet - Institute of Technology
Linköping, Sweden
Supported by
IDA, Linköpings universitet
Vinnova,
project GridModelica
SSF, projects
VISIMOD
and RISE
Local arrangements
Mrs. Bodil Mattsson,
E-mail bodma (at) ida.liu.se
Prof. John R. Gilbert
University of California, Santa Barbara, USA
Abstract:
High-performance computing is being used to understand large data sets that are combinatorial rather than numerical in nature, in applications as diverse as sparse matrices, knowledge discovery, machine learning, search and information retrieval, and computational biology.
Compared to numerical supercomputing, the field of high-performance combinatorial computing is in its infancy. How can combinatorial methods be used by people who are not experts in discrete mathematics?
How can supercomputers be used by people who need to explore huge discrete data sets interactively?
We are building a flexible, scalable interactive environment for high-performance computation on discrete structures that will be used both as a rapid-prototyping tool for exploring and experimenting with different approaches to analysis, and as a scalable system for performing analysis on real, dynamic, discrete data.
Short speaker biography: John Gilbert received a PhD in computer science from Stanford university in 1981. He has been working as researcher at Los Alamos National Laboratory, associate professor for computer science at Cornell University, principal scientist and manager at Xerox Palo Alto Research Center, Since 2002 he is professor for computer science at the University of California, Berkeley. His research interests are in combinatorial scientific computing, geometric algorithms, parallel computing, compiler techniques for high-performance computing, tools and software for computational science and engineering, numerical linear algebra, smart matter and systemic MEMS, and distributed sensing and control.
Prof. David Padua
University of Illinois at Urbana-Champaign, USA
Abstract: This talk will discuss the characteristics and use of hierarchically tiled arrays (HTAs). This class of objects was designed to facilitate parallel programming and the clean representation of algorithms with a high degree of locality. Tiles in an HTA can be distributed across parallel machines or used to represent data ensembles in sequential computations for locality. Array operations can be applied to HTA components in order to compactly represent communication and parallelism. HTAs have been implemented as a class library for MATLAB and C++. The implementation of most of the NAS benchmarks with this library showed that HTAs enable the development of clean, easy to read parallel programs, and the rapid transition from sequential to parallel forms.
Short speaker biography: David Padua is the Donald Biggar Willet professor of computer science at the University of Illinois at Urbana- Champaign, where he has been a faculty member since 1985. At Illinois, he has been Associate Director of the Center for Supercomputing Research and Development, a member of Science Steering Committee of the Center for Simulation of Advanced Rockets, Vice-Chair of the College of Engineering Executive Committee, and a member of the Campus Research Board. He has served as a program committee member, program chair, or general chair for more than 40 conferences and workshops. He served on the editorial board of the IEEE Transactions of Parallel and Distributed Systems and as editor-in-chief of the International Journal of Parallel Programming (IJPP). He is currently Steering Committee Chair of ACM SIGPLAN's Principles and Practice of Parallel Programming and a member of the editorial boards of the Journal of Parallel and Distributed Computing and IJPP. His areas of interest include compilers, machine organization, and parallel computing. He has published more than 130 papers in those areas. He is a fellow of the IEEE.
Prof. David Padua
University of Illinois at Urbana-Champaign, USA
Abstract: The objective of library generators is to automatically produce highly tuned library routines by searching the space of possible algorithms and their implementations. Library generators have proven to be extraordinarily effective in approximating and sometimes s urpassing the performance of manually generated codes at a minute fraction of the cost. Three library generation projects will be discussed in this talk: (1) SPIRAL, a generator of signal processing algorithms based on a database of algebra rules, (2) a modified version of the ATLAS linear algebra generator which replaces the search used by its original implementation with an analytical model which directly generates an optimal version of the library routine, and (3) Genesort, a generator of sorting routines which at execution time adapts to the characteristics of the input data.
Prof. Thomas Rauber
Universität Bayreuth, Germany
Abstract:
Many applications from scientific computing exhibit an inherent modular
structure of cooperating subtasks. For the execution of such applications
on parallel or distributed platforms, a parallel programming model using
multiprocessor tasks is a suitable approach.
Internally, a multiprocessor task can incorporate different kinds of
parallelism or can consist of a hierarchical structure of smaller tasks.
This multi-level parallelism can be exploited to achieve efficient parallel
programs on a specifc platform.
In this talk, we present programming environments and runtime systems which
support multiprocessor task programming based on a flexible specification
of the task structure.
Depending on the specific task structure of an application and the specific
target platform, different functionalities are required.
Statically known module structures may benefit from a priori scheduling
based on runtime predictions for computation and communiation. Other
applcations may require a more dynamic setting.
Short speaker biography: Thomas Rauber is a professor at the University Bayreuth where he holds the chair for parallel and distributed systems. His research interests include runtime libraries and programming environments for task-based executions of parallel and distributed algorithms, performance prediction and locality optimizations. He has published several books and over 70 papers in international journals and conferences.
Prof. Welf Löwe
Växjö University, Sweden
Abstract:
The LOIS/LOFAR projects
aim at constructing a
wide array sensor network in Southern Sweden primarily but not exclusively
to enable research in space astronomy. The sensor network will consist of
more than 13,000 digital receiver nodes, with a combined data rate of more
than 25Tbps, or more than 700 DVD movies per second. The data produced will
be processed by scientific applications, acting as filters on the data
stream.
LOIS/LOFAR and similar systems of scientific applications require huge
computational resources and concepts like the Grid can, maybe, satisfy
these requirements. A computational Grid is a distributed computing
environment with the goal to provide uniform and transparent access to
heterogeneous resources - computational nodes and storage space -
distributed geographically.
While this is a nice idea, the current Grids are inappropriate since sensors
and sensor networks are resources that are difficult to integrate into
these platforms. The reason is that a computational Grid works
batch-job-oriented, while sensor network applications work stream oriented.
In this talk we discuss the requirements of and some approaches to a
Grid-like infrastructure for sensor networks.
Short speaker biography: Welf Löwe has been professor at the Växjö University, Sweden, since 2002, where he holds the chair of software technology. From 1987-1992, he studied computer science at Technical University Dresden, Germany. He received his Ph.D. from Karlsruhe University, Germany, in 1996. After a postdoc visit at the International Computer Science Institute, Berkeley, USA, he returned to Karlsruhe to work as a scientific assistant and later as an assistant professor. The results presented here are embedded in his research on software architectures and platforms for GRID-like parallel systems. Further research topics include component and service based software development, and analysis and visualization of software systems.
Prof. Peter Fritzson
Linköping University, Sweden
Abstract:
The majority of parallel computing applications are used for simulation of
complex engi-neering applications and/or for visualization. To handle their
complexity, there is a need for raising the level of abstraction is
specifying such applications using high level mathematical modeling
techniques, such as the Modelica language and technology.
However, with the increased complexity of modeled systems, it becomes
increasingly important to use today's and tomorrow's parallel hardware
efficiently. Automatic parallelization is convenient, but may also need to
be combined with easy-to-use methods for parallel programming.
The talk gives an overview of the high level Modelica equation-based
modeling language, and three approaches of parallel computing based on
Modelica:
- Automatic Parallelization of Mathematical Models
- Coarse-Grained Explicit Parallelization Using Components (GridModelica)
- Explicit Parallel Programming Constructs in Modelica (NestStepModelica)
Short speaker biography: Peter Fritzson is professor of software technology at Linköping University and Research Director of the Programming Environment Laboratory (PELAB) at the Department of Computer and Information Science, Linköping University, Sweden. He is regarded as one of the leading authorities on the subject of object oriented mathematical modeling languages, and has recently published a book on the matter Principles of Object Oriented Modeling and simulation with Modelica 2.1, Wiley-IEEE Press. He currently holds the positions of chairman of the Scandinavian Simulation Society, secretary of EuroSim, and vice chairman of the Modelica Association, an organization he helped to establish. Professor Fritzson has published ten books and over a hundred scientific papers.
There is no formal registration, but
please send an email to Bodil Mattsson, PELAB,
bodma@ida.liu.se before 2006-06-09 so we
can estimate the number of participants
(for coffee breaks).
You may combine the workshop easily with PARA'06 (June 18-21), to be held in Umeå in northern Sweden.
Please book a room on your own.
Here is information about accommodation in Linköping
by the local tourist information (available in Swedish only, unfortunately).
A relatively cheap accommodation close to the university campus is also offered
by Valla folkhögskola.
For questions regarding local arrangements please contact Mrs. Bodil Mattsson.