DEISA stands for Distributed European Infrastructure for Supercomputing Applications. It is the European GRID initiative. The GRID middleware for DEISA is UNICORE. If you are submitting nbody6++ jobs to the DEISA-GRID, here's the right place to post questions, hints and problems.

1. For a first overview: Please check the DEISA Primer. Section 3 on the DEISA File System might be worth looking at.
2. How to get started:
1. Download the UNICORE client here
2. Get a DFN certificate from your local RA (click here for Heidelberg) and a DEISA_HOME account for which you may apply here (FZ Jülich) -> Benutzer-Accounts für NIC-Projekte (then an Email-Verification follows) -> Put the name of the project: FZJ09HHD. Fill out the online-form and fax the printed version to 02461/61-2810.
3. Importing your DFN certificate for DEISA into your UNICORE client using the Keystore Editor, i.e. Settings->Keystore Editor
4. Import root certificates of all the DEISA sites into UNICORE. You can find them here. It is a package called unicerts.tar.gz. Untar it and import all certificates into UNICORE. Do not care about a few error messages.
5. Go to Settings->User Defaults. Put into the field "URLs for UNICORE Site Server" separated by a comma from the other entries the following url: http://www.fz-juelich.de/unicore/unicoreSites_DEISA.xml. Now all the DEISA Sites should appear in the UNICORE client.
3. Support: The first person responsible for the DEISA project "Einstein" is Marc-André Hermanns. He is responsible for all questions related to enabling, contingent, accounts, etc. There is only one exception: For questions related to UNICORE, the responsible person is Michael Rambadt. He has direct access to the deeper levels of the UNICORE infrastructure and is able to repair malfunctions and problems very quickly, e.g. with the NJS.
4. Benchmarks: Here are some notes for the Benchmarks with nbody6++ (please add your own remarks!):

• The necessary files can be found on Jump in /home8/fzj09hhd/fzj093hd/Benchmarks/
• The file kira.tar.gz contains data sets provided by Derek (named plummer...). These are used by nbody6++ during the benchmark run. The corresponding data set file has to be renamed to dat.10.
• The file in.benchmark.tar.gz contains the input files for nbody6++. The variable TCRIT (the run time) is set to 10 N-body time units. For the larger runs it might not be necessary to integrate for 10 time units and this value may be set to a smaller number. The variable KZ(22) = 8 means that we read a starlab data set.
• The nbody6++ code gives in the output file timings for computation and communication time in seconds at every output interval. This is the line which looks like

Please look at Question 6 in the FAQ of this wiki for the scaling formula and the description of the timing output of Nbody6++.
To find the optimal processor number for a given particle number N, plot at a given time the sums (1) treg + tirr and (2) tmov + tsub + tsub2 against the number of processors np (e.g. double logarithmic, e.g. Log2 for np, Log10 for t). Function (1) decreases and function (2) increases with increasing np. The intersection point lies at the optimal np (A.E., 11.7.07).

1. BenchmarkResults: The results for the Benchmarks are discussed in the BenchmarkResults section. (O.P., 13.02.08)

 Create your own educational wiki!Support CommunityPublic, not yet premium Privacy Policy RSS