| 4-033 | |
| Tools and Infrastructure for Distributed Productions of CMS Monte-Carlo Samples | |
| Names to be provided (for the CMS Collaboration) | We describe the tools CMS has developed over the last two years to produce approximately 20TB of monte carlo data in productions using up to 200 CPUs. Digitising CMS monte carlo events with high luminosity pileup is a complex task, quite unlike the `classic' 1-file-in, 1-file out production. Pileup events must be served to the batch nodes at many times the rate of the signal data, which requires the use of dedicated servers. Signal data is served on demand from mass storage using the AMS to HPSS interface of Objectivity. Output events are written into the Objectivity federation in several streams which are split into separate database files to optimise access during analysis. A Central Data Recording facility interacts with the federation to determine when it can archive data to tape. User federations must be regularly synchronised with the production federation to allow users early access to the data. The AMS Request Redirection Protocol is used both for load balancing and to keep the federation available in the event of server failure. |
Keywords: |
LHC CMS tools production AMS HPSS farm objectivity federation |
| Contact: | Dr. Lucas Taylor |
| Northeastern University / CMS Collaboration | |
| Lucas.Taylor@cern.ch | |