Simulation of binary stars: a testbed for Grid computing
17 November 2004A GASS simulation of Gaia telemetry, corresponding to five years of observations of 1000 astrometric binary stars, has been successfully completed using the GaiaGrid environment.
The Astrometric Binary Star Analysis is one of the shell algorithms selected to be run in the GaiaGrid environment. In order to test this algorithm realistic simulations of telemetry, covering five years of Gaia observations, for about 1000 astrometric binary stars needed to be generated and then ingested into the GDAAS system.
The generation of simulated data in the context of the Gaia mission is under the responsibility of the Gaia Simulation Working Group. The GASS simulator generates realistic telemetry which can be ingested into GDAAS and used to populate the test databases.
In spite of the relatively small number of objects in the astrometric binary star simulation, the generation of the telemetry requires a non-negligible computational effort because several aspects of the simulation are not related to the number of objects but rather to the simulated time span. GASS was deployed in the GaiaGrid environment in order to exploit the computational power it provides. In the case of GASS this deployment is particularly effective because the design of the simulation allows it to be completely split into several smaller pieces that can be simultaneously sent to different nodes of GaiaGrid, taking full advantage of the distributed computing environment provided.
The simulations were run in GaiaGrid using the Gridassist workflow tool during a week in October 2004. After some initial testing and fine-tuning, 183 independent jobs were launched to the GaiaGrid environment, covering the full period of five years of simulation. These jobs were automatically distributed to the GaiaGrid nodes by Gridassist and the results were collected in a central repository located at CESCA (Centre de Supercomputació de Catalunya), accessible through the Grid. In summary:
- The simulation was run in 23 nodes distributed in 8 institutes of 5 countries (see figure of GaiaGrid sites)
- A total of 3.8 million CPU seconds were used for the tasks
- A total of 16.5 GB of data were produced and automatically transferred to the central repository at CESCA