VO Technologies
O1.1 High Performance Computing (Invited)
- M. Livny (Univ. of Wisconsin)
No abstract provided.
O1.2 Rapid Development for Distributed Computing, with Implications for the Virtual Observatory (Invited)
- Michael Noble (Harvard-Smithsonian Center for Astrophysics)
Compute nodes will require enabling technologies to participate in a virtual observatory (VO), yet much of the available GRID software exhibits the classic buy-in problem: a) it is distributed in relatively large packages that require regular updating as new versions are deployed, and platform-specific binaries for each CPU architecture, b) involves steep learning curves, and c) can require significant institutional commitment to install, utilize, and maintain.
These factors create a formidable entry barrier for metacomputing newcomers, a class into which a large majority of the astronomy community currently falls. Recent work, however, shows that combining Java tuplespaces with network class loaders can greatly simplify the configuration and management of distributed computations within heterogeneous networks. When added to the semantic clarity of tuplespace programming, such an approach could shrink the buy-in cost of VO participation considerably, at the very least in terms of the up-front costs needed to merely dabble in metacomputing, as well as simplify the iterative redeployment of experimental infrastructure and services that will prove necessary as the VO evolves.
To bolster this argument we outline an architecture which shows promise as a means of accessing diverse data and services through a virtual metacomputer interface. Early prototypes suggest that the approach offers a number of attractive VO-relevant features: (1) loosely couples service provider nodes, accessor nodes, and users across networks of all scales; (2) permits dynamic joining and leaving of service providers and accessors; (3) hides much of the complexity of distributed programming behind a clean and simple interface; (4) provides intrinsic scalability and fault tolerance, and comparatively simple replication; and (5) requires low institutional or individual buy-in, relative to other GRID tookits, for initial VO participation.
O1.3 STAR: Building an Observational GRID
- Alasdair Allan, Tim Naylor (University of Exeter) Iain Steele, Dave Carter, Jason Etherton, Chris Mottram (Liverpool John Moores University)
The eSTAR Project is a programme to build a prototype robotic telescope network to design and test the infrastructure and software which could be used in larger scale projects. The network consists of a number of autonomous telescopes, and associated rapid data reduction pipelines, connected together using GLOBUS middleware. Intelligent agents carry out resource discovery, submit observing requests, and analyse the reduced data returned by the telescope nodes. The agents are capable of carrying out data mining and cross-correlation tasks using online catalogues and databases and, if necessary, requesting follow-up observations from the telescope nodes. We discuss the design and implications of the eSTAR software and its implications with respect to the GRID.
O1.4 AstroComp: A Web Portal for High-Performance Astrophysical Computing on a Grid of Supercomputers
- Paola Di Matteo, Paolo Miocchi (Univ. of Rome) Vincenzo Antonuccio-Delogu, Ugo Becciani (Astroph. Obs. of Catania) Roberto Capuzzo Dolcetta (Univ. of Rome) Alessandro Costa (Astroph. Obs. of Catania) Vittorio Rosato (ENEA - Rome)
AstroComp is a project (initially funded by the Italian National Research Council, CNR) aiming at creating a portal that permits to handle and use high-performance numerical tools for Astrophysics, on a grid of supercomputers. The main motivation of the project is to construct a portal, which allows to set up a repository of computational codes and common databases, making them available and enjoyable, with a user-friendly graphical web interface, to the entire national (and international) community. AstroComp will allow the scientific community to benefit by the use of many different numerical tools implemented on high performance computing (HPC) resources, both for theoretical astrophysics and cosmology and for the storage and analysis of astronomical data, without the need of specific training, know-how and experience either in computational techniques or in database construction and management methods.
I will illustrate some examples of practical utilization of the present version of the AstroComp portal in the framework of the numerical simulations of globular clusters dynamics. I will show how to handle the various aspects related to the performing and managing of a typical N-body simulation.
A prototype of the portal can be visited at http://www.astrocomp.it/
