IT-protoDUNE coordination (Single Phase and Double Phase)

Europe/Zurich
31/S-028 (CERN)

31/S-028

CERN

30
Show room on map
Description
Coordination: JIRA, Minutes, etc. https://twiki.cern.ch/twiki/bin/viewauth/ProDUNEIT/WebHome
    • 15:00 15:05
      Coordination update 5m
      Speaker: Xavier Espinal Curull (CERN)
    • 15:20 16:00
      Status reports (Tour-de-table) 40m
      Speaker: All

      Input from Geoff:

      From the online side we are awaiting the return of Marc Collignon.  The data transfer links also need to be configured and tested.

      Wireless is supposed to be installed throughout EHN1 this week.   We have GPN connections in the control room and electronics room barracks that we use in place of wireless when possible.  The GPN connections for the DAQ barracks are awaiting a switch installation.

       

      Input from Stuart:

      One potential topic…  In the Data Challenge we are now running the first step is to copy Monte Carlo data files (typically 8GB in size) from EOS to another EOS directory acting as the “drop box”.  This is to mimic the eventual copy of files from the DAQ in EHN1 into that same drop box.  To make these EOS to EOS transfers we use 3rd party xrdcp (for this example we using a single stream, but elsewhere we use 4 streams).  We also do similar EOS to EOS copies later in the data management chain, but for now let’s concentrate on this first step.  In this exercise, we’re typically doing 10 parallel xrdcp copies for this “first step” in the data path, but that is followed by an equal number of parallel paths to (1) yet another EOS location, (2) Castor, and (3) Fermilab dCache (gridftp for the latter); so we are keeping EOS busy.

      We see a wide variation in the transfer rates, almost “bi-modal”.  See the attachment.

      Our hypothesis – on which we are asking for comments – is that some of the transfers are Meyrin to Meyrin, but others may be Meyrin to Wigner (and vice versa).  But we are speculating.

      The question for discussion at the meeting:  Does this hypothesis make any sense?  Are there alternative explanations?  Can we “work around” this to do better?

       

      Input from Marc (while on vacation):

      For your information, the CDR dedicated link (40G) from NP04 to the computer center is routing with static routes to avoid other traffic. From now, only the IP address range 128.142 is routed via this link. We need to add some IP ranges in 188.184 (to be checked) to cover completely the computer center and Wigner.

      This could be an explanation because this traffic to Wigner will use the GPN default route. I’m not at Cern today and unfortunately couldn’t check this. Don’t hesitate to give me more details about the sources and destinations for this EOS traffic and I’ll have a look as soon as possible.