VAMDC Project (2009 - 2012)

Reference Details
González, Juan A., Mendoza, Claudio, Núñez, Luis A., Witthoeft, Michael C. and García, Javier (2009), "XSTAR: a data-intensive and CPU-intensive astrophysical application within the VAMDC", Proceedings of the Second EELA-2 Conference: p. 317.

Abstract:
The Virtual Atomic and Molecular Data Centre (VAMDC) is a recently launched initiative to deploy an advanced cyber-infrastructure for distributed atomic and molecular (A&M) database services. A&M data are heavily used in a variety of research and industrial fields such as astrophysics, atmospheric physics, fusion, environmental sciences, laser technology and lighting. VAMDC involves research groups from the European Union (Austria, France, Germany, Italy, Sweden, UK), the Russian Federation, Serbia and Venezuela. Data activities, software developments and computations at the Venezuelan node are currently being supported by EELA-2. The integration of existing A&M databases within VAMDC must necessarily address at least three perspectives: data producers’; data evaluators’ and data users’. In the past ten years, we have worked in a computational astrophysics project with groups from Mons University (Belgium), NASA-GSFC (USA) and IVIC (Venezuela), among others, which include both data producers and data users. This project has established efficient channels not only to handle in detail the A&M data requirements of a particular astrophysical application, namely the XSTAR spectral modeling code [1], but also to generate a quality A&M database which can be used beyond the context of the actual application by a wider data-user community. Thus, the porting of data-intensive applications to a distributed e-science environment such as VAMDC implies other issues besides code performance, i.e. data management and integration, which are the main topic of the present report. Spectral modeling is the proverbial astrophysical application, which uses extensive A&M databases to resolve the physical conditions of an astronomical emitting body, e.g. its plasma temperature and density and chemical abundances, by generating a synthetic spectrum comparable with the observed. The plasma model equations must take into account, say, the excitation mechanisms, ionization fractions, statistical equilibrium and radiative transfer. Due to the spectacular advancement in both terrestrial and satellite-borne telescopes, observed spectra now cover the whole electromagnetic range: radio, microwaves, infrared, visible, ultraviolet, X-rays and -rays, with interregional correlations growing in demand. XSTAR, in particular, is frequently used to study the high-energy phenomena displayed by compact-body systems, e.g. active galactic nuclei and X-ray binaries, by modeling photoionized plasmas in the X-ray spectral region. In regular conditions, a successful XSTAR model must be run many times in order to identify its sensitivities to variations of input parameters and, therefore, it may be a very CPU-intensive procedure. The upscaling of XSTAR to the cyber-infrastructure requires two complementary aspects: (i) the transcription of an atomic flat-file database onto a structured SQL schema that can be easily modified, maintained, exported and integrated within the new virtual-data-center initiatives; and (ii) code gridification to increase performance and turnover, particularly when used in the multi-model mode. The atomic database of XSTAR is currently being improved by including new massive and accurate data sets for K-vacancy levels (level energies, transition wavelength, radiative and Auger rates and high-energy photoionization cross sections) in metals with atomic number 7 ≤ Z ≤ 28 (see [2] and references therein). With respect to database redesign, the idea is to construct a self-consistent data set. This means that, not only do we combine several different data sets together, we also need to combine potentially overlapping datasets in an intelligent way. The database needs to store atomic data with well-defined data types which share the same parameter space. Such a structure allows for data sets to interrelate with each other. The database structure must be able to describe the data it contains (fundamental data types, data types, parameter types and sources), and it must be flexible enough to store multiple data types in a common framework. Furthermore, a library of functions must be built to act on the database where consistency in function names and actions is very important in order to make script writing easier and less prone to bugs. Multi-model XSTAR gridification, on the other hand, has required the development of scripts that dynamically generate gLite job-description-language (JDL) files [3] that can be run on the grid as multi-node parametric jobs, where the final model summary, model tables and log file are being produced locally upon job return to the UserInterface. For the latter, we are considering both the grid portal and virtual machine options. [1] Bautista, M. A., Kallman, T. R. 2001, ApJS, 134, 139. [2] Witthoeft, M. C., Bautista, M. A., Mendoza, C., Kallman, T. R., Palmeri, P., Quinet, P. 2008, ApJS, 179, 542. [3] JDL Attributes Specification, EGEE-JRA1-TEC-590869-JDL-Attributes-v0-9, https://edms.cern.ch/document/590869/1

Webpage Link  Webpage Link
[ Back ]
Copyright © 2018 VAMDC Project (2009 - 2012). All Rights Reserved.
Joomla! is Free Software released under the GNU/GPL License.