Where should you store and analyse you private Ntuples?

In December 1998, the Cern PIAF facility was switched off. Delphi users used PIAF mainly for two reasons:
  • to interactively analyse their Ntuples with PAW
  • to store their Ntuple files on the PIAF disks
    This page describes the replacement of this service.

    Disk space

    The user Ntuples now reside on the Delphi disk server shd01, filling a volume of 100 Gb. This machine is connected to the dxplus and shiftdelphi clusters by GigaBit Ethernet, the fastest network at Cern. Therefore, to efficiently read your Ntuples, you have to connect to one of these services. To access your files, use the environment variable $DELPHI_PAW. It is expanded to "/piaf/delphi/u/username", and can be used like this:
  • from Paw, open your file with:
           HI/FILE 99 $DELPHI_PAW/myfile.rz
  • to read your Hbook file in a Fortran program:
           ln -s $DELPHI_PAW/myfile.rz .
           CALL HROPEN(LUN,'MYHIST','myfile.rz','N',LRECL,ISTAT)
  • using FTP:
           ftp dxplus.cern.ch << EOFTP                 
              get $DELPHI_PAW/myfile.rz                
  • As the data disks are as readable any other, you can also:
           cd $DELPHI_PAW
           ls -lrt $DELPHI_PAW/*.rz

    But these data disks are mounted read-only!
    To write on these disks, use the RFIO commands:

  • to copy a file to this Piaf area:
           rfcp myfile.rz shd01:$DELPHI_PAW/myfile.rz
  • to remove an old file:
           rfrm shd01:$DELPHI_PAW/myoldfile.rz
  • to rename a file:
           rfrename shd01:$DELPHI_PAW/myfile.rz shd01:$DELPHI_PAW/newfile.rz
  • to create a subdirectory:
           rfmkdir shd01:$DELPHI_PAW/183
  • to list your files:
           rfdir shd01:$DELPHI_PAW
    (or from dxplus : ls -l $DELPHI_PAW)
    The "rfcp" command replaces "topiaf" !
    Note the syntax, with the "shd01:" before "$DELPHI_PAW"

    Interactive PAW sessions:

    On dxplus a special command pawbatch has been created for running CPU intensive PAW sessions. pawbatch runs PAW semi-interactively, in a special LSF queue. In this way, PAW runs on the dxplus node which is least busy, and without any CPU limit. The drawback: you need to wait some 15 seconds after typing pawbatch before it really starts. If you do not intend to consume a lot of CPU, a regular PAW session on dxplus will of course work as well.

    Some miscellaneous points:

  • the commands topiaf and frompiaf are now obsolete
  • if you compile your Fortran selection function in PAW with the native compiler (as is recommended for speed), beware that the native compiler on dxplus is stricter than on the old PIAF! You may discover some bugs in your code...
    As always, questions, complaints, remarks: