DKRZ system overview

The DKRZ hardware and software infrastructure is optimally adapted to accomplish the needs of complex climate modelling research. This section (including the the graphic below) should give the user a schematic overview on the DKRZ systems. (for a more task or work flow based description please look here.)

HPC-system blizzard

Blizzard, which is the main compute source at DKRZ, is an IBM Power6 cluster, running with UNIX operating system AIX. It provides some common tools like emacs, vim, eclipse, unix tools (e.g. grep, diff, awk, sed, etc.), version control systems (subversion, git, git-svn etc) etc. to work with the source code of a model. The user has full access to the parallel file system GPFS and his/her data on the File Systems HOME,  WORK and SCRATCH.

Login node blizzard.dkrz.de

• working with source code: emacs, vim, eclipse, unix tools (e.g. grep, diff, awk, sed, etc.), version control systems (subversion, git, git-svn etc) - Blizzard
  
• building of binaries: (g)make, IBM XL compiler, gcc - Compilers, Libs & Tools, Fortran runtime environment
  
• submission, monitoring and cancelling of Batch Jobs: TWS LoadLeveler
• none memory- and time-consuming data pre- and post-processing: cdo, afterburner, netCDF, ncview, etc.
• connection to HPSS tape archive: pftp
  

Special interactive node p249

• performance of development and test runs: IBM PE - Interactive Node
  
• Debugging: ddt, dbx, gdb, marmot, nmon
• accessible only by request to beratung@dkrz.de
  

Compute nodes p001-p246 (batch mode)

• performance of production runs: IBM PE
  
• routine data pre- and post-processing: afterburner, CDO, NCO, etc.
  
• Program Analysis of applications: Scalasca, VampirTrace, hpccount, libhpm, libmpitrace, tprof, gprof (non-shared node usage recommended)

HPC-node  passat

• for interactive work login to passat.dkrz.de and use it like blizzard  login node
  
• used for serial queue "serialp7", see  LoadLeveler queue limits
• connection to GPFS filesystems  HPSS tape archive: pftp
  
  

Linux cluster lizard

lizard (like blizzard) is connected to GPFS and the nodes will also be usable by LoadLeveler in the near future - use it for:

• time- and memory intensive data pre- and post-processing: CDO, NCO, netCDF, afterburner, ncview, tar, gzip, etc.
  
• data analysis and simple visualization: MATLAB, mathematica, R, Scilab, NCL, GrADS, FERRET, IDL, etc.
  
• submission, monitoring and cancelling of batch jobs: TWS LoadLeveler (not implemented yet)
  
• connection to HPSS tape archive: pftp
  
  

Visualization cluster halo

The Vis Server Halo is also connected to GPFS and is designed especially for:

• high-level and 3-D data visualization, animations: Avizo Green, NCL, IDL
  
  

HPSS tape archive

• backup, storage, and long-term archiving of data: pftp from within DKRZ/ZMAW network
• sftp, gridftp read only access from elsewhere
  
  

Linux cluster tornado 

• working with source code: emacs, vim, eclipse, unix tools (e.g. grep, diff, awk, sed, etc.), version control systems (subversion, git, git-svn etc)
• building of binaries: (g)make, Intel, PGI, NAG, SUN and gcc compiler
  

• performance of low-resolution or short-term experiments: OpenMPI, SGE
  
• data processing and simple visualisation: CDO, afterburner, NCO, NCL, GrADS, R, IDL, FERRET, ncview, netCDF, etc
• tornado will be decommissioned in July 2012