P14 Reprocessing: Site Certification

Each site must prove that it is able to run d0reco and produce results identical to those found at FNAL. We agreed with Harry Melanson that the following procedure would suffice for certifying the remote sites:
  1. Run recocert on fully reconstructed events from FNAL farm for use as standard set. Note that Harry has run recocert on other FNAL p14.05 output and he considers it "certified".

    This only need be run on thumbnails. It is not necessary to run this on DSTs.

  2. Copy the thumbnails (including metadata and logs) of the certification run (176852) from remote sites to d0mino. Each station has a recocert area for this in /sam/remote_merge. E.g. ccin2p3 would copy their certification thumbnails to: /sam/remote_merge/ccin2p3/recocert
  3. The metadata should be checked by one of the experts.
  4. Recocert should then be run against these thumbnails and overlayed on the reference set from the FNAL farms.
  5. If any differences show up in the overlays, then we will ask Harry and/or Suyong to examine the plots and declare them fit/unfit. It was agreed that on the first passes of this procedure we WILL ask them to examine the plots regardless of how inconsequential any differences may seem.
  6. List of selected events from remote site should be compared to Marco's master list. The match should be exact.
  7. No difference in the certification plots or blessing by Harry/Suyong, and an exact match on the selected event lists gives a remote site a green light to proceed with production.
There are two things in here which you might question. First, use of thumbnails rather than DSTs. Harry's contention is that any signiifcant differences should show up in the thumbnails since any useful physics object is in there. That makes our life a lot easier since we have limited temporary storage space for DSTs. A second point which might concern you is the comparison using fully reconstructed output from FNAL and selected output from the remote sites. Harry points out that although there will be differences in things like electron quality, the more basic quanitites such as numbers of tracks and hits are driven by normal background and not the few "interesting" tracks in an event. Consequently, things like track and hit distributions should not differ much between selected and unselected samples. So I think we have our marching orders. Remote sites should copy their thumbnails from the certification run to d0mino. They should check their selected event lists from this run agains Marco's master list ( I think both gridka and ccin2p3 hav already done this). Phil should run recocert on the thumbnails in sam for run 176852 as a reference. When the remote sites have notified Phil that their thumbnails are on d0mino he should run recocert on them and overlay against the refence set. We will then ask Harry/Suyong to bless/curse the output.
Mike Diesburg, Daniel Wicke, 28. Oct. 2003. Last update 31th Oct. 2003.
Diesburg@fnal.gov
Daniel.Wicke@physik.uni-wuppertal.de