|
Документ взят из кэша поисковой машины. Адрес
оригинального документа
: http://www.adass.org/adass/proceedings/adass96/devriesc.html
Дата изменения: Tue Jun 23 21:15:21 1998 Дата индексирования: Tue Oct 2 02:47:59 2012 Кодировка: Поисковые слова: п п п п п п п п п п п п п п р п р п р п р п р п р п р п р п р п р п р п р п |
Next: Three-dimensional Data Analysis in IRAF and ZODIAC+
Previous: The DRAO Export Software Package
Up: Software Systems
Table of Contents - Index - PS reprint - PDF reprint
C. P. de Vries
Space Research Organization Netherlands (SRON),
Sorbonnelaan 2, 3584 CA Utrecht, Netherlands,
E-mail: C.deVries@sron.ruu.nl
A critical requirement for data analysis systems used for massive routine data processing is that the system is able to deliver processed data of controlled quality in an automatic fashion. In order to control quality of data, it is necessary to have the ability to trace the heritage of all data products. This means recording all parameters of all steps which lead to the establishment of the final data products (Figure 1). To check processing status and initiate subsequent processing steps, one should easily be able to generate overviews of all available data and intermediary products based on data descriptions and processing heritage.
The SRON-HeaD (SRON High Energy Astrophysics department Data analysis) system has been developed to fulfill these requirements, and is based on earlier experiences with the CGRO-Comptel data analysis system (de Vries 1995).
Figure:
A processing pipeline is a sequence of tasks with several input/output files and parameters, which pass data from one task to the next.
Heritage of any file (e.g., file 9) can be established by recording all input/output files and parameters of all steps and by uniquely identifying each task, requiring thorough configuration control.
Original PostScript figure (48kB).
The following basic requirements were defined:
The core of the system is the recording of data descriptions and data heritage in the database, where this information may be queried via user interfaces or the routine processing pipeline to start new jobs (Figure 2). The dataset heritage consists of actual processing parameter values and the software configuration used. Since data processing parameters are available either in the FITS headers or from input parameter files used by the generating programs, these parameters can be recorded after actual data files have been created. This means that no connection to a database is required during data processing and that external packages can easily be incorporated into the system. In addition, externally generated (FITS) data files can easily be imported into the system, as well.
Figure: Top level functional breakdown of the HeaD system.
Original PostScript figure (48kB).
The basic processing module is a ``task'' or program executable, called from a script or ``job,'' which may also call other tasks. The script defines the control flow and communication between tasks within a job.
The job script may prompt the user for task parameter input. Actual parameters are passed to tasks via IRAF format parameter files. Automatic recording of processing parameters is done for each output file at the end of job processing.
An Oracle client-server database architecture serves as the central database system (RDBMS), which holds the data catalogue and heritage, processing parameters description and software configuration. Software configuration at the system level is maintained through use of the RCS system, and use of this system is enforced by the appropriate user interfaces.
The Tcl/Tk package plus extensions (TclX,OraTcl) is used to create the windows-based user interface (Figure 3). On-line help is available by means of a Tcl/Tk HTML viewer. Direct selection of data files for input to the appropriate user interfaces can be made by selecting from lists resulting from the execution of SQL procedures. These can be taken from a library of procedures which base selections on a variety of data descriptions, heritage, or processing status. In addition, Oracle-Forms can be used for direct database queries.
Figure: HeaD user interface.
Original PostScript figure (2.1MB).
Data processing ``jobs'' can be initiated manually by explicitly entering task parameters for specific tasks, or automatically through manual or automatic marking of individual datasets or dataset types for further processing. In that case, special database tables define the processing flow.
A data access layer (DAL) is available which separates the actual scientific code from basic data I/O, allowing for greater system portability. This layer is partially composed of the FITSIO library, modified to allow for communication with the archive system, and specially developed routines.
Currently the system contains specially developed data processing programs as well as tasks taken from general packages like FTOOLS, IDL, SAOimage, etc. The system has been initially implemented on Sun (SunOs, Solaris) and HP (HP-UX) systems.
More information can be found on HEAD Home Page.
de Vries, C. P. 1994, in Astronomical Data Analysis Software and Systems III, ASP Conf. Ser., Vol. 61, eds. D. R. Crabtree, R. J. Hanisch, & J. Barnes (San Francisco, ASP), 399
Next: Three-dimensional Data Analysis in IRAF and ZODIAC+
Previous: The DRAO Export Software Package
Up: Software Systems
Table of Contents - Index - PS reprint - PDF reprint