The brochure of the John von Neumann Institute for Computing is available in English and in German. It can be ordered at the NIC secretariat (nic@fz-juelich.de).

deutsche Broschüre (pdf)   |  English brochure (pdf)

Intro- duction	Scientific Computing	Astro- physics	Elementary Particles	Multi- particles	Polymers	Chemistry	Earth, En- vironment	Other Fields

    Scientific Computing

• "Scientific Computing"
• IBM Supercomputer Jump in Jülich
• Automatic Cassette Robot
• Response Time Monitor
• Massively Parallel APE Architectures
• JuNet and its Connections to the Rest of the World
• VIOLA: a Testbed for Innovative Network Technologies and Grid Applications
• Grid Computing
• Computational Science and Engineering
• Automatic Performance Analysis of Parallel Programs with KOJAK (Kit for Objective Judgement and Automatic Knowledge-based Detection of Bottlenecks)
• Computational Steering of Jump Applications with VISIT

"Scientific Computing"

For many years scientific computing on high-performance machines has played an important role in large research institutions world-wide. Research Center Jülich entered the field of supercomputing in 1983 when it installed the CRAY X-MP computer, the first vector computer of this performance class in Europe. Scientific computing has since been constantly further developed towards high-end computing, simulation and modeling by exploring new algorithms and methods, extending the range of applications as well as by integrating innovative computing resources.

The flagship of the computer resources currently available for NIC at Jülich is the IBM supercomputer "Jump" that was installed in January 2004. Additionally, a vector computer and various cluster systems are available. Magnetic disks and magnetic cassette robots provide a capacity of 2.2 petabytes for data storage.

For applications in the field of theoretical high energy physics, NIC develops and operates parallel computers with SIMD architecture at DESY in Zeuthen. This architecture has proved particularly efficient in the simulation of quantum chromodynamics.

In order to offer NIC users the greatest possible benefit from the computer resources available, research and development are indispensable for integrating the high-end computers into the production environment, for example by developing new monitoring systems, and for connecting them to high-speed networks. The further development of communication technologies, for example within national collaborations like the VIOLA testbed, is, therefore, of special importance.

NIC users receive intensive support from project-accompanying consulting services, and they profit from an exchange of experience during courses and user workshops. NIC-ZAM offers a three-step concept for user support. The help desk solves the small, daily problems. Specialists provide support for methods and program optimization. Additionally, each project is assigned an advisor, with whom it may enter into a long-term scientific partnership.

The well-qualified support of the various NIC projects requires both a thorough understanding of the simulation methodology and the provision of tools for the cost-effective use of the systems and for easy, yet secure access. Thus, the research and development activities of NIC- ZAM concern computational sciences, computer science, and applied mathematics. These activities include the development of grid environments, tools for the optimization of parallel programs (KOJAK), or for visualization and steering of simulations.

NIC-ZAM strives to maintain its position as a world-class center for high-end computing. Therefore a primary objective is to observe and evaluate new computer architectures and future systems - be it clusters with fast interconnects or leadership-class systems for highest performance.

(Rüdiger Esser, NIC-ZAM, Jülich)

IBM Supercomputer Jump in Jülich

The supercomputer, an IBM p690 cluster nicknamed Jump (Jülich Multi Processor), has 41 nodes with 32 Power4+ processors with a clock frequency of 1.7 GHz and a shared memory of 128 Gbytes. All nodes are connected via the quick connection network "High Performance Switch". The complete system reaches a peak performance of 8.9 teraflops. Users can submit jobs which use several hundred processors and up to 5 terabyte memory.

Automatic Cassette Robot

Three silos of this magnetic cassette robot with a storage capacity of 2.2 petabytes are available for the supercomputer in Jülich.

Response Time Monitor

The response time behavior is an important factor for evaluating the performance of interactive systems. In order to measure this behavior and to detect bottleneck situations at an early stage, a response time monitor (RTM) was developed and installed to supervise the central interactive computer systems in Jülich. An RTM agent simulates a "standard user's" interactive session by a sequence of commands, program requests and "pauses for reflection". The measured response times and other parameters (e.g. system utilization, number of active users) are transmitted to a WWW server, which makes the data available as an HTML document for the Internet.

(Wolfgang Gürich, NIC-ZAM, Jülich)

Massively Parallel APE Architectures

For applications in theoretical particle physics, DESY has installed a specialized massively parallel supercomputer, APEmille, in Zeuthen. This machine comprises 1104 processors distributed over 8 crates and connected by a fast 3-dimensional communication network. The machine achieves a peak performance of 583 GFlops. Users performing simulations of lattice quantum chromodynamics reach a very high efficiency of 50% or more on the machine, which, together with its stable operation, makes the APEmille the workhorse for lattice physicists in Germany. The computer time on the APEmille machine is allocated by NIC through its peer review board.

The APEmille machine was developed in a collaboration between the Istituto Nazionale di Fisica Nucleare (INFN) in Italy and DESY. The success of this machine and the future computing requirements as evaluated by a European ECFA panel convinced the INFN and DESY to adopt the ambitious plan of developing a successor machine, apeNEXT, that should be able to reach the 10 TFlops performance range, which would also satisfy the computing needs of the Lattice Forum (LatFor) community in Germany. The University of Paris Sud joined the collaboration and other institutions, such as the University of Bielefeld, contributed to the project as well. The development of the processor and all the other hardware and most of the software components of the apeNEXT machine has been successfully completed. By the beginning of 2005 a prototype machine with 1.6 TFlops peak performance is expected to run application codes. This forms the basis for even larger installations that can operate in the multi-teraflops regime.

(Karl Jansen, Dirk Pleiter, Hubert Simma, NIC DESY-Zeuthen)

JuNet and its Connections to the Rest of the World

ZAM’s task is to provide efficient communication systems according to the respective state of the art. Today, the largest fraction of the campus network of the Research Center, JuNet, is a centrally managed, switched 100 and 1000 Mbit/s Ethernet network relying on fiber-optical connections. It is complemented by a rapidly growing wireless network.

Apart from ISDN, which is primarily used for access to JuNet from the private sphere, the Research Center’s world-wide network integration is effected via the gigabit science network (G-WiN) of the DFN Association (Association for the Promotion of the German Research Network), currently with a 622 Mbit/s link. As a partner in the VIOLA project, the Research Center is participating in preparations for the X-WiN, the next generation science network.

In addition, ZAM and two universities in the Jülich-Aachen region jointly operate a network between their respective campuses, which is based on leased dark-fiber lines. It enables co- operations with demanding communication requirements between these institutions. Currently, the fibers are used for the Research Center’s access to the G-WiN point-of- presence in Aachen and for the research projects DEISA and VIOLA.

Prior to the introduction of innovative communication techniques, ZAM carries out pilot implementations and beta tests of new equipment generations.

(Thomas Eickermann, NIC-ZAM, Jülich)

VIOLA: a Testbed for Innovative Network Technologies and Grid Applications

In addition to the ever growing demand for bandwidth on the Internet, the emerging new paradigm of grid computing generates the demand for a new quality of network services. Grid computing provides technology for seamless access to and use of distributed resources like supercomputers, huge storage facilities or experimental devices, and thus allows scientists and engineers to work and collaborate in unprecedented ways. Grid applications typically require flexible on-demand provision of high communication bandwidth often in combination with quality-of-service guarantees. Today’s data networks do not provide sufficient support for these new demands.

The BMBF-funded VIOLA project (Vertically Integrated Optical testbed for Large Applications) addresses these issues and serves as a pilot for X-WiN, the next-generation science network in Germany. A consortium of 6 universities and research centers and 3 industrial partners, coordinated by the DFN Association, is setting up and operating a national optical testbed (initially with multiple 10 Gbit/s Ethernet and SDH links) in the Aachen- Jülich-Cologne-Bonn region. An extension to Bavaria and a link to the European research network GÉANT are in preparation. The project will test and deploy new networking components and architectures and develop software for dynamic bandwidth allocation.

The focus of ZAM in VIOLA is the further development of grid applications to enable them to make optimal use of the new capabilities of the network. Such applications are the distributed simulation of pollutant dispersion in the soil (in co-operation with ICG-IV) or the distributed collaborative visualization of huge atmospheric data sets. UNICORE is used as the grid middleware and will be enhanced to support simultaneous allocation and use of distributed resources including the required communication bandwidth.

(Thomas Eickermann, NIC-ZAM, Jülich)

Grid Computing

Grid computing is an evolving key technology that will enable scientists and engineers in research and industry to solve challenging problems, master complex environments, and collaborate in unprecedented ways. Grids will integrate distributed computing resources, data produced by scientific instruments - like tomographs, accelerators, satellites, or telescopes -, data created through simulations and stored in archives or databases, and visualization media through high speed networks. The resulting knowledge environment will allow virtual organizations to be formed dynamically and research and development to be pursued in novel ways with increased efficiency. In short, grids are enablers for e-science.

ZAM has a proven track record in grid computing. It led the development of UNICORE - Uniform Interface to Computing Resources - a vertically integrated middleware with the following unique features. Users can create and execute complex workflows in a seamless, secure and intuitive way on a wide range of systems at any of the sites participating in a grid. UNICORE translates the tasks that the users specify through a graphical interface into a sequence of system-specific commands for the selected target system and controls their execution. Users no longer have to master the internals of different systems nor the conventions at different sites. UNICORE has built-in end-to-end security based on X.509 certificates, the accepted standard for grids, which provide authentication, single sign-on, and encryption where required. UNICORE respects the autonomy and security policy of participation sites and integrates into proven computing center operations.

The UNICORE software is available royalty-free as an Open Source under BSD license. ZAM supports academic users of UNICORE and coordinates its future development and deployment. UNICORE has been and is the basis for several European projects that enhanced the original software and contributed to the world-wide recognition of UNICORE. For example, the GRIP project (Grid Interoperability Project) demonstrated that UNICORE and Globus complement each other naturally. UNICORE jobs can execute seamlessly on resources managed by either system; the users will see no difference, and applications do not have to be changed to use the Globus Toolkit. One of UNICORE’s outstanding features is its extensibility to support a wide variety of applications and combine them in unprecedented ways. The OpenMolGRID project has the objective of accelerating drug design using grid technology. Based on properties of chemical compounds that are available in different databases new molecules are constructed in silico and their likely properties are computed. Only the most promising ones are synthesized and tested in the laboratory. To automate this process UNICORE creates a complex work flow (screenshot above) based on high-level specifications from the scientist. The result may be several thousand computations using different algorithms and involving distributed computers and databases (figure in the middle). The chemist views the most promising candidates (screenshot below). Grid computing will only succeed if viable standards are created. The Open Grid Services Architecture (OGSA), as defined by the Global Grid Forum (GGF), is the prime candidate for an overall framework within which specific standards are being specified by standards organizations like OASIS and W3C. ZAM is actively involved both in GGF and OASIS. The EU UniGrids project - Uniform Interface to Grid Services - ensures the evolution of UNICORE towards an OGSA-compliant system.

(Dietmar Erwin, NIC-ZAM, Jülich)

Computational Science and Engineering

Computational science and engineering aims at the simulation of problems that are otherwise difficult or even impossible to investigate. Although different research fields, like quantum chemistry, solid state physics or molecular dynamics, have developed specific methodologies over time, they face similar technical/computational challenges. Thus they benefit substantially from interdisciplinary cooperation between the natural and engineering sciences, computer science and applied mathematics. NIC-ZAM participates in method development, in particular for HPC systems, in close cooperation with leading research groups; the focus of this development is primarily on physics and chemistry due to their relevance for the NIC user community.

Some major research areas investigated at NIC-ZAM are shown in the figure. The topics range from large size but rather low accuracy many-particle problems to very small size but complex elementary particle physics calculations: Many-particle dynamics is concerned with the time evolution of systems containing up to 100 million particles. Due to the physical properties it is necessary to describe processes on a time scale of 10^-15 seconds, whereas the whole simulation time might go up to 10^-3 seconds leading to approx. 10¹² time steps per calculation! Hence, the accuracy of the model has to be restricted for the sake of acceptable computing times.

Chemical reactions like catalytic reactions on solids are difficult problems in industrial chemistry, but of considerable practical relevance. A realistic model must resort to quantum mechanics, i.e. the approximate solution of Schrödinger's famous equation. Fortunately, there are methods for these kinds of problems, like density functional theory, which deliver useful insights at relatively low computational cost. However, if additional effects, such as excited states of the molecules, become important for the reactions, less approximate quantum mechanical methods must be used, but at the price of increasing computing times.

Moving to the smallest physical entities, namely the atomic nucleus, its components or other elementary particles, one finally ends up in lattice quantum field theory, where even the present teraflop supercomputers are not powerful enough to give conclusive answers on the subatomic structure of matter.

Whether the focus is on accessible system size or the accuracy of the model, all these different topics have in common that they rely on the latest computer architectures, tools for the optimization and/or parallelization of the research codes, as well as powerful and efficient (mathematical) algorithms in order to perform state-of-the-art research. Thus an important part of NIC-ZAM’s mission is to foster computer simulations in science and technology by method development in key scientific areas, by supporting the efficient usage of advanced computer systems, as well as by training scientists and students.

(Bernd Körfgen, NIC-ZAM, Jülich)

Automatic Performance Analysis of Parallel Programs with KOJAK
(Kit for Objective Judgement and Automatic Knowledge-based detection of bottlenecks)

The application of parallel high-performance computers for the investigation of scientific problems serves to execute complex simulation within an acceptable time. The efficient use of existing resources is a prerequisite for the rapid execution of individual programs and for optimizing the overall throughput. The typical process of performance analysis consists in the repeated execution of program instrumentation, program execution with performance measurement, and analysis of the performance data by the user until the inefficient program parts are recognized. All three steps require great experience in handling the analysis tools and well-founded knowledge concerning possible performance bottlenecks. The aim of developing KOJAK is to automate this process as much as possible (see diagram). Based on a database with detection rules, an analysis component automatically locates, classifies, and assesses potential performance bottlenecks and reports them to the user sorted according to their negative impact on the performance of the program. In addition, if necessary, this information enables a guided and therefore more efficient manual analysis with conventional tools.

(Bernd Mohr, NIC-ZAM, Jülich)

Computational Steering of Jump Applications with VISIT

The well-known problem solving cycle of computer simulation, post-processing, visualizing and re-adjusting of simulation parameters can be significantly improved and accelerated by computational steering, an interactive connection between a simulation program and a visualization. It enables the user to control intermediate results and to change simulation parameters immediately and thus use both the computational resources and his own time more efficiently.

For this purpose, the VISIT library (Visualization Interface Toolkit) is being developed at the Central Institute for Applied Mathematics. It supports the development of interactive simulations and provides functions for establishing a connection between a simulation and a visualization, exchanging data and eventually shutting down the connection again. VISIT provides support for parallel applications and includes a directory service (SEAP) for resource discovery. Currently supported visualization systems are AVS/Express, IDL, VTK, and Perl/Tk. In the EU UniGrids project a grid services interface for VISIT will be developed, which allows integration into the UNICORE grid environment.

(Wolfgang Frings, NIC-ZAM, Jülich)

Intro- duction	Scientific Computing	Astro- physics	Elementary Particles	Multi- particles	Polymers	Chemistry	Earth, En- vironment	Other Fields