Transmenu powered by JoomlArt.com - Mambo Joomla Professional Templates Club

Lattice calculations have become an essential tool to realize and exploit the full potential of the European investment in accelerators and detectors. At present, they are the only means to extract exact non perturbative predictions of QCD from first principles with controlled systematic errors. They set the stage for key experiments at the research infrastructures and are crucial for the interpretation of the anticipated experimental results.

Accomplishments over the last 5 years have established the methodology and laid the groundwork for lattice QCD calculations. Since the cost of full QCD computations grows with a large inverse power of the quark mass, initial calculations were restricted to relatively heavy quarks. In order for lattice calculations to reach the needed accuracy requested by the experiments, simulations at physical quark masses and close to the continuum limit are inevitable. To reach this goal, Teraflop/s-scale computing facilities, a unified programming environment, and the pooling of resources are required.

The purpose of this work package is to provide the hardware and software infrastructure necessary for world class lattice QCD research, and to utilize this infrastructure for scientific discovery. It has proven more cost effective to build dedicated computers rather than to make use of general purpose machines. We plan to assemble and operate a computer consisting of a 3-dimensional array of Enhanced Cell BE processors, QPACE (for QCD parallel computer using cell technology), with an aggregate peak performance of more than 416 (208) Teraflop/s in single (double) precision, optimized for lattice QCD calculations. To achieve Petaflop/s performance, and to utilize the full capacity of the next generation of multi-core processors, which will become available in 2009, a more powerful communication network needs to be developed. In preparation for a Petaflop/s system, we plan to build a prototype machine, whose key elements are a high-bandwidth, low-latency network and the Intel 16 core Nehalem processor. Among the applicants are members of the Italian-German-French consortium that designed and built the apeNEXTcomputer, and some of the architects of the QCDOC machine.

For certain thermodynamics calculations, with moderate demand on computer memory and communication, high-end graphic cards (GPUs) promise to be a powerful alternative to massively parallel architectures. The Wuppertal consortium plans to install a cluster consisting of 32 nodes with 2 GPUs each, with a sustained performance of 2 Teraflop/s in single precision.

Altogether, this goal can only be achieved on a cooperative European level. All the efforts can be coordinated in a Joint Research Activity (JRA) in Lattice QCD.

This JRA will pioneer the design of massively parallel capability computers based on multi-core processors. The QPACE development project includes various Universities and Research Laboratories in Germany and Italy, all of which are participants in this activity, as well as the IBM Böblingen Research Laboratory as industrial partner. The development costs of QPACE, which are substantial, will be defrayed by IBM and Deutsche Forschungsgemeinschaft (DFG) through the Sonderforschungsbereich/Transregio TR55 “Hadron Physics from Lattice QCD”. While QPACE is based on current technology, the prototype machine will utilize the next generation of network hardware and multi-core processors. The advantage of custom built machines is that they provide an unprecedented price-performance ratio.

The QPACE project is well advanced, and the full machine will be installed in Q2. The task then is to exploit it in a real production environment, and to utilize it for research in lattice QCD. The prototype computer will be installed by Q18. Upon successful operation of it, a larger system is envisaged. The effort of making such new technology usable for research and, eventually, commercial applications, is often underestimated. This JRA will bridge the gap.

It will also pioneer the use of high-end graphic cards, like Nvidia 8800, for high-power computing (HPC). It is expected that the gathered experience will be very useful for other groups and other applications with large demand of computer power.

The project is organized into 11 main topics. Some of them cover the assembling, testing and operating of the prototype machine and of the graphic cards, while the others concern the physics problems to be studied:

-          Hardware deployment;

-          Machine and application software;

-          Generation of background field configurations;

-          Chiral dynamics;

-          Light hadron spectrum;

-          Hadron structure;

-          Heavy quark physics;

-          Fundamental questions and other directions;

-          Bulk thermodynamics;

-          In-medium hadrons and screening;

-          Universality, Chirality and the hot QCD vacuum.

These topics are subdivided into several tasks, which are assigned to and shared among the various participants according to their expertise and manpower. This is shown in the following table.

The QPACE machine will enable calculation of the physical observables at pion masses down to 140 MeV and lattice spacings down to 0.05 fm on spatial volumes larger or equal to (4 fm)³ with high statistics. Combined with chiral perturbation theory, controlled extrapolation in pion mass, lattice volume and lattice spacing will be possible.

• Operational prototype of a GPU-based computer (delivery month from start date: 12),
• Operational 416 Teraflop/s QPACE computer (delivery month from start date:6)
• Operational prototype of a next generation multi-core processor machine (delivery month from start date:18)
• Operational software and libraries for the next generation machine (delivery month from start date:24),
• Physics results and publications (delivery month from start date:30);

A large number of physics results and publications, following the tasks defined here above, are anticipated.

Lattice QCD, like other areas of science, is limited by computer performance. In the long run it will demand sustained speeds of Petaflop/s or more. This presents a greater challenge than the recent step to Teraflop/s, because massive parallelism must be delivered at an affordable price, and many codes running on present systems will not scale up to run efficiently on ten to hundred times more processors or cores. Hardware that is tailored to the application, and/or serendipitous use of cheap commodity parts developed for some other purpose, will be needed to keep machine costs down, and software will need to be re-engineered to use it efficiently. These factors are driving us towards a diversity of architectures, international facilities and community code bases. There is scope for innovative solutions, which gives Europe the opportunity to re-enter the high-performance computing arena. We believe that a multi-core processor based parallel system provides a particularly promising solution.

This JRA will strengthen existing capabilities and the development of new high-performance computing competence in Europe. It will enable the JSC-Jülich and ZIB-Berlin Computing Centres, as well as DESY, to provide truly European service to the wider hadron physics community. Furthermore, our combined efforts in testing, writing and disseminating optimized scientific codes will contribute significantly to the quality and performance of the European computational infrastructure. Participation in this project will also have a significant training benefit.

Last, but not least, this joint effort will give us complete control over the systematic errors inherent in the numerical calculations, thus providing a unique opportunity for a fruitful synergy between lattice gauge theory and the research infrastructures CERN, DESY, FZ-Jülich, GSI, JLab, LNF-Frascati and MAMI, as well as JSC and ZIB.