SCS home page
|
Description of the fieldWe seek to discover and quantify the intrinsic computation in dynamical complex systems through investigation of the computational structure of such systems. We treat these systems effectively as a virtual parallel computer, for example by associating the evolution of spatial patterns with the performance of parallel computation. Spatially extended model systems, such as cellular automata, are used to address the intrinsic parallel computational aspects, and to derive generic algorithms (called natural solvers).An important aspect is the study of the mapping of complex systems onto high performance parallel and distributed computing systems. The research can be regarded as contributing to the foundations of a theory of parallel computation in complex systems. We strongly believe in experimental validation of all theoretical work. Therefore, based on expertise gained in previous years, we will continue to study implicit and explicit parallelism in applications of large-scale dynamical complex systems. Research challenges and MysteriesA common denominator in the next generation scientific computing applications is the understanding of multi-scale phenomena, where systems are studied over large temporal and spatial scales. The simulation of these natural phenomena requires an ever increasing computer performance. Although computer performance as such still doubles approximately every year, it is our strong belief that the development of algorithms for modern computer architectures stays behind. One of the biggest challenges is, therefore, to develop completely new algorithmic approaches that support efficient modeling of natural phenomena and -at the same time- support efficient distributed simulation. Hence we need to boost the computational power instead of the computer power.One way to approach this is to look closely to the way nature itself performs computation. This is largely an unexplored research field. Our approach is to use the concept of interacting virtual particles whose dynamics give rise to complex behavior, by using Cellular Automata as a compute metaphor for modeling and distributed simulation. AlgorithmicAs specific instances of Cellular Automata we study Lattice Gas Automata and the Lattice Boltzmann model. Although these models have been around for a decade, they were mainly studied from a theoretical physics point of view. Our interest is to study them from a computational science point of view, to apply them to real-life natural phenomena and to compare them with real experiments. Within the community this is acknowledged as being of a very high relevance. We are on the front of the second wave of interest in discrete particle models, where the computational aspects and the modeling abilities are the main research questions.MethodologicalNow that a computational speed of a Teraflop/s is reality, the international arena of High Performance Computing is preparing for the next big leap, the Petaflop/s. It is expected that the first computer to break this barrier will be the GRAPE6, a hybrid computer specifically designed to execute particle models with long range interactions. We participate in an international collaboration, led by P. Hut of the Institute of Advanced Studies in Princeton, and J. Makino from the University of Tokyo, that aims to build GRAPE6. We investigate mapping of (virtual) particle models to new parallel hybrid architectures. Both architectural design and parallelization of particle models for these designs guides our research.By the decoupling of processes from processors in a novel discrete event simulation approach, where processes can freely migrate over a processor system, we can create a highly portable and efficient runtime support system. ThemesWithin the SCS the research is organized around four themes:
|
| Most recent modification: Wed 15 Jan 2003 by dick at science.uva.nl | 10
hits in May 2012 |
