| prev toc next |
We present a numerical gravitational solver, based upon a multigrid algorithm. The multigrid technique allows to solve accurately the Poisson equation using the relaxation method implemented over a hierarchy of computational meshes with increasing resolution. This allows to improve strongly the convergency and the accuracy of the method without increasing the computational weight and ensuring a linear scalability with the number of cells. A local refining mesh startegy (Adaptive Mesh Refinement - AMR) is exploited, which allows to have high resolution computations exactly where it is needed. Lower resolution (and therefore low computational effort) is instead provided wherever this is sufficient for the correct description of the system. This makes this solver particularly suitable for cosmological simulations, in which an extremely large dynamical range is required. The open source DAGH library is used for the management of the adaptive grid. The code is parallelized using the standard MPI library. The gravitational solver will be released open source to the scientific community.
The growth of the World Wide Web and the present availability of very fast networks, as fast as the computer's internal links, can turn into reality the computer science dream of disintegrating the machine across the net into a series of special purpose appliances like computing elements, storage elements, meta-data catalogues, resource brokers, information indexes, etc. The infrastructure of such a world-sized computer is often termed as
Future NASA space science missions will require the use of computers capable of executing billions of instructions per second while supporting billions of bytes of memory with a power budget of only a few tens of watts available to the computing system. NASA's Office of Space Science is currently evaluating proposals demonstrating that commercial-off-the-shelf (COTS)-based high performance flight computer technology is mature enough for space validation tests in 2006. NASA's goal is to have parallel-computers in space with more than 100 computing nodes each of which uses low power components, is radiation tolerant, and is very reliable (99.999% availability; a down time of just 26.3 minutes over a 5-year mission lifetime). This Office of Space Science program (NRA 03-OSS-02) could possibly provide technology that will enable future NASA space missions in the next decade to have access to computational power with total throughput values that are one to two orders of magnitude greater than state-of-the-art radiation-hardened flight computers can provide today. I will first discuss the astronomical scientific use case for high-performance computers for space. Next, I will describe the hardware and software technological challenges of building high-performance parallel-processing computers in space. Finally, drawing upon recent advances in the commercial Linux-server market and current research in fault-tolerant parallel-computing, I will describe some possible solutions which might overcome many, if not all, of these technological challenges within the next few years.
We present some preliminary results of the analysis of long-term light curve sequences of magnetically active close binaries by means of a numerical parallel code made available through the Astrocomp web portal. The code looks for the best values of the photometric parameters and evaluates their confidence intervals for eclipsing binaries with cool spots on their surface. Specifically, cool spots produce distortions of the shape of the light curves that may cause systematic errors in the determination of the luminosity ratio, the fractionary radii of the component stars and the inclination of the orbital plane. Our method of analysis reduce the effects of spots on the determination of such parameters by simultaneously fitting a long-term sequence of data along which the spots' coverage and distribution show remarkable changes thus allowing us to correct for their perturbation (cf. Rodon\`o, Lanza \& Becciani 2001, A\&A 371, 174).
The large amount of computational work required by our approach is managed by means of a parallel code based on MPI. The load balance of the computation is achieved by estimating the time required to analyse a given light curve and distributing subsets of light curves to the available processors in order to get similar overall computation times.
The code has been made available through a web-based user-friendly interface called Astrocomp. It allows a registered remote user to run the parallel code on a set of high performance computing resources.
Astrocomp is a project developed by the INAF-Astrophysical Observatory of Catania, University of Roma La Sapienza and ENEA (http://www.astrocomp.it).
The operation of novel phased array based radio telescopes requires vast amount of computational power for the processing of observed data. In the LOFAR telescope, the central on-line and off-line systems will use HPC resources provided by a 1000+ nodes cluster computer. In order the profit from the enormous computational power if this machine a application development framework is built. This framework support the transformation of the current single-node data reduction applications to parallel equivalents that can execute on large clusters. In the online systems, real-time signal processing pipelines can be executed. Furthermore, the complexity of controlling many concurrently running observations and analysisese on an inhomogenious hardware set is tackled. In this presentation the architecture of the framework is presented and the applicability to astronomical observations and LOFAR in particular is discussed.
| prev toc next |