It is foreseen that ATLAS will take data with Heavy Ion (initially Pb-Pb) collisions for approximately one month per year, starting in late 2008 [2-8]. Here we assume that the Heavy Ion data-taking period will last 106 seconds (50000 seconds/day for 20 days) each year. The trigger rate is effectively limited by the available bandwidth between the HLT and Tier-0 buffers (320 MB/s) and the event size (5 MB) to ~50 Hz. We take this as a reference number for the purpose of this document.
The Computing Model for Heavy Ion data is not the same as for p-p data. A significantly reduced Event Data representation is assumed, and the simulation and processing will be tailored to reflect this. The assumptions about data sizes and processing times are given in Table 2-3. Note that the simulated datasets will be a mixture of central, peripheral and `model' events.
There is no need foreseen for two reprocessings of the data in the year in which it is taken. instead, a small fraction will be processed during the data taking for control and monitoring, and then the full processing performed during the gap between p-p processing at the Tier-0. While it makes a marginal impact on the computing model, it is assumed that only a subset of Tier-1s (perhaps only two) and a corresponding subset of Tier-2s will house and process the heavy ions data.
Another possibility is that Heavy Ion data are distributed only to the subset of Tier-1s hosting the communities of physicists most interested in those data, and processed only there. This model necessitates a study of the available network bandwidth to those computing centres and significantly increased resources to be made available in those centres, although of course the total amount of required resources would remain the same as in the "simplest" model above.
Copyright © CERN 2005