Minutes of the splinter on Beowulf and Archive WP4-WP5

1 November - Lorentz center Leiden

People present:

K. Begeman, E. Bertin, E. Deul, A. Grado, M. Neeser, M. Pavlov, E. Puddu,

R. Rengelink, R. Silvotti, M. Snigula

A. Grado presented the result of the tests performed on the PartenoBEO cluster in Naples. Following the action list Q2 2002, OAC has created a tool, called astro-bench, that runs application astronomical benchmarks on a cluster and is able to profile the network, CPU and disks usage. We all agree to distribute the tool to all the national data centers in order to compare the hardware available at the different sites. The software is available, what is missing is a graphical interface to show the profiling results. In November Astro-bench will be completed and distributed to all the AW partners. (AI Grado)

The results shown in the talk demonstrate that a cluster with eight slave nodes plus a master already meets the processing speed of 1Mpix/s in the case of a 8 CCD mosaic. The presentation triggered the discussion about the cluster configuration and the archive implementation. For the archive we have basically three possibility, the first is an external independent storage system attached only to the master, the second is to have the storage system in the master/masters and the third is to have the archive distributed on the nodes. Tests and discussion are necessary in order to identify the best solution in terms of scalability, availability, reliability and costs.

For the end of Q4 2003 all the data centers must have a Beowulf cluster with a scalable archive that will grown accordingly to the data center needs. Basically, the next actions are to distribute the Astro-Bench tool to the data centers (Grado) to subsequently compare the performance of the different clusters available (AI Snigula, Begeman, Deul, Bertin, Grado).

There will be a dedicated section on the topic to compare and discuss the results in the next Astro-Wise meeting.