Minutes26Feb2010 RAC core team minutes, February 26, 2010
Core Team Participation
Jim Cochran (Analysis Support Manager),
Michael Ernst (U.S. Facilities Manager)
Kaushik De (Operations)
Armen Vartapetian (U.S. Operations Deputy)
Note Concerning Minutes
Minutes of the monthly meetings of the RAC will be finally approved at monthly meetings.
Minutes of the core team meetings may be approved at the weekly core team meeting.
Urgent need to free up disk space at Tier 2s (Kaushik)
A few days ago the MWT2 had reached 96% utilization of DataDisk. This had triggered an examination of the current practice of replicating 100% of data and MC AODs at MWT2 and also NET2. The Operations team proposed reducing future data and MC replication to 50% at MWT2 and NET2. Already replicated datasets would remain in place. The proposed reduction would leave US ATLAS with two full copies of the AODs plus four 50% copies.
The RAC approved this proposal.
Kaushik noted that DataDisk at MWT2 was now down to 20% utilization (Details of disk space usage at all US ATLAS Tier 2s are attached and were circulated by Armen before the meeting). He did not know why so much data had been deleted and intended to ask the MWT2 what had happened.
Michael and others commented that reliable usage information was not available and would be vital when making future decisions like this. Kaushik reported that ATLAS was expected to introduce usage-based automated deletion of datasets in about three months from now.
Kaushik noted that the CREM recently made a decision to delete all but one copy of the RAW data and that this had caused jobs to fail at BNL.
Discussion of needs for disk space reporting and "political" processes to allocate disk space (Richard)
In principle ATLAS controls the pledged disk space, but in practice, the degree of replication and the placement of data are currently decided by US ATLAS. ATLAS has honored all US requests for replication and placement. US ATLAS has not requested any deletions to date, partly because there is not enough monitoring to know whether users are accessing data.
Richard pointed out that xrootd provides detailed logs of file access and Michael confirmed the dcache writes detailed “billing logs”. All such logs are too detailed to be used without further processing. Kaushik noted that DQ2 logs DQ2get calls and Panda also tracks dataset polularity. It is not clear to what extent it is safe to act (delete) on the basis of the DQ2 or Panda information.
Richard raised the longer-term issue of making ATLAS resource allocation tractable and scalable. This could be achieved by allocating quotas to physics and performance groups within which the group could manage resources. Achieving this would require ATLAS-wide agreement and some technical development. Perhaps RAC members should promote this concept within ATLAS, taking care to take a collaborative approach.
Michael agreed that this was, in essence, the right way to go. There was no dissent.
- 2/26/2010: Kaushik, Find out why the MWT2 DataDisk usage decreased to 20%
- 2/19/2010: Richard, Announce the RAC and its Wiki to usatlas-l (Completed)
- 2/19/2010: Richard, Propose RAC talk at US ATLAS PI Meeting (Talk Scheduled)
- 2/12/2010: All (but especially Kevin/Jim/Ian), Propose the parts of the activity list above that are your responsibility and identify any activities that do not appear to be an appropriate task for any of the current RAC members. (Ongoing)
- 2/12/2010: Richard Non-urgent simulation tasks - consult with physics groups on the best way to establish two-way communication with the RAC to meet this and other needs (Ongoing)
About This Site
Please note that this site is a content mirror of the BNL US ATLAS TWiki. To edit the content of this page, click the Edit this page button at the top of the page and log in with your US ATLAS computing account name and password.
(9.5K) | RichardMount
, 01 Mar 2010 - 17:21 | US ATLAS Storage on February 26 2010