Clusters of galaxies are tracers of the large-scale distribution of matter in the Universe, so cosmological parameters can be derived from a statistical analysis of cluster surveys. While clusters can be identified in either optical or x-ray surveys, the reliability can be greatly improved by correlating the two. The necessary algorithms have been developed at the Max-Planck-Institut für extraterrestrische Physik, but carrying them out requires large amounts of computational resources. To enable a reliable all-sky survey of galaxy clusters, grid computing techniques are being developed in the framework of the German Astronomy Community Grid, a research project for scientific work supporting e-Science and Grid middleware, with participation by all important German astronomy institutes and grid-specific research groups from computer science and supercomputing centers. The core calculations are carried out by a Fortran 90 program called Clusterfinder, but for this program to take advantage of the Grid, solutions had to be developed for a large number of diverse issues. The research demonstration will present in real-time the approaches taken in the following areas.
The German astronomy grid has dozens of computing resources that are very heterogeneous. The first challenge is getting a running version of the software onto each of dozens of hosts. Transport and versioning of source code is done primarily with Subversion. Since some systems do not have Subversion, the system switches to gsiscp if necessary. Setting up the environments for compilation, testing, and production is done with a flexible combination of site-local information (e.g. using environment modules) and centrally-maintained information (again using the Subversion repository).
A typical run will involve choosing a patch of sky and setting physics and computational parameters. For setting parameters and submitting a job, a GUI based on GridSphere is being developed.
To relieve the user from keeping track of which hosts are able to run the Clusterfinder and which of these are currently available, submission is made through an instance of GridWay.
As input data the Clusterfinder needs the coordinates and brightness of galaxies from the Sloan Digital Sky Survey (SDSS) and the coordinates of x-ray detections from the ROSAT All-Sky Survey (RASS), both of which are available online. Clusterfinder obtains the data needed for the requested patch of sky from these databases using OGSA-DAI.
A production calculation requires a complex workflow. This starts with production of likelihood maps based on SDSS and RASS independent of each other, then combining these maps, and finally extracting a catalog of the clusters found. Other tasks are dividing the sky into different patches for the calculation and then recombining the results, or making parameters scans. The workflows are coded and run using either Gnu make or the Process Coordinator begin developed by the Planck project.
Although the primary product of Clusterfinder is the catalog of clusters, the
researcher would also like to examine and explore the results visually, both
to assure himself that the result is correct and also to discover new features.
This is currently offered by IDL routines integrated into the workflow, the
output of which is shown in Figure 1.
Figure 1: Visualization
using IDL of a galaxy cluster discovered by the Clusterfinder program. The
likelihood map in the upper left is from the visible galaxies alone, in the
lower right from the x-rays alone, in the upper right from the combination of
the two. In the lower left is the superposition of an image in visible light and
the contours of the x-ray emission.
Acknowledgment: This work was funded by the German Federal Ministry of Education and Research.