NL-T1 site configuration

Update at Nov 2010: Originally (4-5 years ago) we were using only the NIKHEF CE as the SARA CE was very small. This Tier1 had the peculiarity that the CE and SEs were in different institutions. So we called historically everything NIKHEF. Now SARA CE is as large or larger than NIKHEF's CE and therefore we decided to use:
  • SARA for reco/stripping
  • NIKHEF for analysis (and merging). We could get disk storage at NIKHEF that would move the T0D1 storage of NIKHEF-SARA to NIKHEF, and only RAW, RDST and M-DSTs would remain at SARA as NIKHEF has no tapes... but this is additional work and we never found the time to work on that...
Look at the storage mapping for NIKHEF and SARA and you will see that the "close SEs" to SARA are RAW and RDST only. this trick makes it to split the two activities...

Proposal by Philippe 16.11.2010:

1. Rename NIKHEF-xxx as SARA-xxx and make NIKHEF-xxx an alias of SARA-xxx

2. Use SARA in jobs such that new files are labeled as SARA-xxx (i.e. update Tier1-SE definition), but keep NIKHEF as close SEs to both SARA and NIKHEF for access to legacy files.

Then we have two options: 1. Rename files in the LFC (probably necessary for RAW anyway) 2. Wait until there are no longer files at the NIKHEF SEs (considering we still have files on -disk and -tape, this may take a while wink

At least we have time for proceeding with the SE renaming in the LFC... Anyway we have to go through this exercise if/when we move to DPM at NIKHEF for T0D1 storage.

Dec. 2010: problems due to fair share: we do not have enough CPU slots to run the reprocessing at SARA After discussion with Jeff and Ron in April we decided to share the load between the (effectively two different) sites in a deterministic way in order to not overload the network between the two buildings. We were assured at the time that there was enough computing power at SARA and thus we assigned the reconstruction jobs to SARA. But now it seems that SARA only has 300 CPU slots! As far as we are concerned, NL-T1 is a single Tier1 and should behave so. Would it be possible to increase our fair-share at SARA and possibly decrease the fair share at NIKHEF in order to keep the same total share at NL-T1? In our computing model we have to assign a site for (re-)processing each run. Now these runs have been transferred to SARA and assigned to SARA. One thing we can try and do is to assign the CEs to in the CS...

Jeff: The problem is not the split between Nikhef and SARA of CEs, it's the split between Nikhef and SARA of the configuration of the fair shares. I don't mean here how many percent FS, but things like over what amount of time is the FS computed and is the FS calculation absolute or relative. LHCb is running many more jobs at Nikhef than at SARA, with no sawtooth behaviour, and the shares in terms of % is not that different between Nikhef and SARA IIRC. Let us take this up internally, this should not require action on your part, LHCb!

-- ElisaLanciotti - 17-Nov-2010

Edit | Attach | Watch | Print version | History: r2 < r1 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r2 - 2010-12-07 - unknown
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Main All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback