Minutes of MC+SAM+Release Meeting
            =================================

              June 20, Wednesday, 1:00-3:00pm
                       9th Circle

Scribe: Qizhong Li

Attend: Iain Bertram, Mike Diesburg, Dave Evans, Alan Jonckheere, 
        Qizhong Li, Lee Lueking, Serban Protopopescu, Vicky White

1) Metadata from mc_runjob for SAM
  Problem: 
           Iain reported that the current remote farm runs p07.00.05a
           MC programs and p08.10.00 (non-opt) D0reco and reco_analyze.
           But the metadata produced by mc_runjob for reco_analyze does
           not match SAM's requirement.
  Solution:
	   Dave and Lee have discussed details of the metadata format.
           Dave will produce an mc_runjob to handle p07.00.05a MC exe's
           and p08.10.00 d0reco and reco_analyze, which will produce
           the metadata according to Lee's requirement.
           Then Iain will test it and report the test results to the
           production group (Boaz) and the simulation group (Qizhong
           and Serban).

2) Cardfiles
  Problem: 
           Alan reported that the updated cardfiles package was not in 
           the production release until p08.11.00 and p09. A patch to
           minitarfile has been made for the remote farms, which uses
           p07.00.05a MC. In general, the physics groups may often need 
           their most recently designed cardfile for requesting new MC 
           events. These new cardfiles may not be in the production 
           release the remote farm uses.
  Solution:
           Cardfiles should not be a part of the production releases.
           The mc_runjob should use cvs tag number for cardfiles rather
           then use the production release number. Dave will modify
           the mc_runjob for this.
           Alan will make a new minitarfile of the cardfiles package 
           for the remote farms whenever someone requests a newer
           cardfile, which has a newer cvs tag than the tag on the
           farm. So the remote farm can get the new cardfiles from
           the minitar area.

3) Serban's requests on generating single tau events.
  Problem:
           Serban has described his specific request to generate
           100,000 single tau events on the farm. These events need
           the following special processes.
            a) run MCSingle_x  (to generate ASCII file of 4-momenta)
	    b) run MCbyHand.x  (for special needs)
	    c) run D0gstar2.x
	    d) run D0Sim_x      (with different RCPs)
            e) run D0reco_x     (with a set of different RCPs to turn off
                                most of packages and use different vertex)
            f) run RecoAnalyze_x  (with a set of different RCPs)

  No decision is made on this one. Since this involves changes for the
  farm operation, the production group (Boaz and Mike) need to decide
  how they want to process.

  A possible solution is suggested:
          - Serban will run a) and b) locally on d0mino to generate
            the 100k events at the generator level. Let's call the 
            file, which Serban will produce, the "Input File".
	  - Serban will produce correct metadata for the 'Input File".
          - Put the "Input File" into SAM. (Vicky pointed out that
            SAM will have to be modified to be able handle Serban's
            metadata. That will take some time.)
          - Serban will provide a set of special RCPs to be used for
            D0Sim, D0reco and Reco_analyze. (This seems need to go
            through RECO Czar.)
          - With RECO Czar's permision/request, Alan will build a
            production release including all these special RCPs.
          - Then the remote farm takes the "Input File" from SAM and
            run it through mc_runjob. (Iain pointed out that this
            need to change the farm production procedure to handle
            the special process. e.g. to get proper metadata for
            the Input and Output files, to record the number of events
            in the Input and output files, to get all these special
            RCPs for different EXE's, modify mc_runjob... 
            Iain said, this is not something can happen right now. 
            They need time to test it and make it work. Of course, 
            all these are under assumption that Boaz agrees with this.)

4) In addition to the above three problems, there are several reports.
  
   - Lee and Vicky reported that they are working with Dave together
     on a system, which can record the metadata from mc_runjob to
     a database in SAM. The project is just started. No estimate on
     when it can be finished.

   - Alan pointed out that we need to record all kind of information, 
     so that we can reproduce the events five years later if needed.

   - Iain reported that Reco_analyze crashes all the time. It can not
     run for more than 200 events.

   - Mike reported how the local farm is operated. It seems it has
     a lot of patches to fix crashes. So the local farm does not use
     the exact version of the production release.

   - Serban would like to see that the farm can run generators with a 
     selector. The selector code will be in cvs and test release,
     but not in the production release.