ATLAS Tier 1 Resource Requirements Timetable for 2006/2007/2008

Last Updated 1.11.2007: Start to add 2008 plans and clarify site offers compared to experiment requirements.

Updated 21.08.2007: Add placeholders for the M5 and M6 Cosmics runs for end October and end December.

Updated 26.07.2007: Add details on the M4 Cosmics run scheduled for 23 August to 2 September.

Updated 25.06.2007: Split off 2006 plans into a separate linked page and remove LHC engineering run.

Updated 14.05.2007: Drop RAL from ESD pairing with ASGC and TRIUMF mirroring to SARA.

Updated 23.02.2007: Add plans for data distribution tests in Feb/March and May, full scale dress rehearsal from July to October and MC event generation for the rest of 2007.

Updated 15.01.2007: Move the Tier0 export tests from 15 Jan to new preliminary date of end Feb.

Updated 17.11.2006: Revise (downwards esp. disk space) MC requirements for first half of 2007.

Updated 31.10.2006: add Monte-Carlo plans up to mid-2007 and January 2007 Tier-0 and export exercise.

Updated 4.8.2006 to add repeat of June T0 to T1 export for last half of September.

Updated 27.07.2006 to extend Atllas T0-T1 data export through August.

Updated 29 June to extend Atlas T0-T1 data export to the end of July.

Updated 12 June to add request for sites to report data rates to tape.

Updated 2 June to add 3 week rampup plans leading to the 19 June tests.

Updated 8 May to reflect date change of T0-T1 exercise from last 3 weeks of June to 19 June to 7 July. This has not yet been reflected to monthly site plans.

Following the Mumbai meeting in January 2006 Atlas Tier-1 shares were defined as follows and used for the 2006 site resource estimates:

These were refined after the December 2006 GDB meeting and with new site pledges to be used for the 2007/8 site resource estimates and which add up to 100% of the requirements. These percentages are used to calculate the per site resource requirement spreadsheets:

For 2008/9 the Tier-1 site offers in cpu+disk (we take the average) exceed the requirements by about 11%:

Relationships between ATLAS Tier1 Sites

Given the different amount of resources that each Tier1 site will provide to ATLAS, the following 'pairing' has been defined for the mirroring of (primarily) ESD data that is produced by the reprocessing at the different sites showing the percentages of the total ESD that are to be exchanged between the mirrored groups.

BNL will in addition take a complete copy of the ESD.

Distribution of activities over 2006

Distribution of activities over 2007/8

ATLAS Distribution of activities over 2007

The assumptions behind the 2007/8 data distribution tests are a 200 Hz trigger rate, a raw event size of 1.6 MB, an ESD event size of 1 MB and an AOD event size of 0.1 MB. Atlas want to demonstrate that sites can run stably at the resulting instantaneous rates but also assume an accelerator efficiency of 50000 seconds per 24 hour day so define the success metric of these tests as reaching a daily average of 5/8 of the instantaneous rates. Tier1 sites share raw data and first pass ESD data from the Tier 0 according to their pledges but each site gets another share of ESD data according to the relationships above and in addition BNL will get a complete copy of the ESD. The model is that raw data is to be stored on tape at the tier 1 and the ESD on permanent disk until replaced by later reconstruction. In addition each Tier 1 will receive a complete copy of the Tier 0 AOD to be redistributed to their associated Tier 2 sites. For the tests raw and ESD data can be recycled as the sites wish but sites should have a disk buffer of 3 days worth of AOD data and use this to redistribute AODs to their T2 sites before deleting the AOD. The target data rates out of Tier 0 will hence be raw at 320 MB/s (average of 185 MB/s), ESD at 508 MB/s (average of 294 MB/s) and AOD at 200 MB/s (average of 116 MB/s). The Atlas planning Twiki for this activity is at https://twiki.cern.ch/twiki/bin/view/Atlas/TierZero20071