Hi Erik, thanks a lot for the script. I will let you know what I get.
I will circulate the files necessary for simfactory as soon as I'm sure that everything is working properly. One "problem" I have now is that Janus requires users to compile codes on a compile node which is only accessible via the login node. I don't know if it is possible to configure simfactory in order to be able to access the compile node via the login node. If not, I can also ask the administrators if there is a way to access the compile node directly, without having to first login on the login node. At the moment I compiled the code directly on Janus instead of doing it remotely from my laptop.
Cheers, Bruno
On Sep 20, 2012, at 6:03 PM, Erik Schnetter wrote:
Bruno
Congratulations! Feel free to add the Janus option lists etc. to Simfactory.
I've use a TOV star in the past for benchmarks. I attach a Perl script that creates parameter files.
There are a few lines in the beginning where you choose some parameter ranges. You probably want to set hydro to 1, mincores to the number of cores per node, and adjust maxcores as you see fit. You should also define the number of threads that should be used; this list may contain only a single number if you want. Start by setting run_list to (0).
This will create parameter files that use approximately the amount of memory requested (500 MB per core), and will run approximately for the amount of time requested (300 s).
The script will create a number of parameter files. You would submit them all, running each once. You would look at carpet::timing..asc of each simulation, and look at column 22 there ("total_grid_points_per_second"). This is the benchmark results, and excludes startup and initialisation time.
-erik
On Thu, Sep 20, 2012 at 6:55 PM, Bruno Giacomazzo bruno.giacomazzo@jila.colorado.edu wrote: Hi, I have installed the EinsteinToolkit (Lovelace) on the Janus supercomputer here at the University of Colorado Boulder (https://www.rc.colorado.edu/resources/janus). Most of the testsuites pass (see also summary.log attached). The ones that give more problems are those from QuasiLocalMeasures which all fail because of segfault and I will try to understand why.
I'd like to know if there is a "standard" benchmark parameter file that I can use to test the performance of GRHydro on Janus. Ideally it would be something simple (e.g., a TOV, so that I could use a very similar parfile to compare also GRHydro vs Whisky) and that you have used on other machines. I'm in particular interested in testing its weak scaling and possibly also its strong scaling.
Thanks, Bruno
Dr. Bruno Giacomazzo JILA - University of Colorado 440 UCB Boulder, CO 80309 USA
Tel. : +1 303-492-5170 Fax : +1 303-492-5235 email : bruno.giacomazzo@jila.colorado.edu web: http://www.brunogiacomazzo.org
There are only 10 types of people in the world: Those who understand binary, and those who don't
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
-- Erik Schnetter schnetter@cct.lsu.edu http://www.perimeterinstitute.ca/personal/eschnetter/
<bench-create>
Dr. Bruno Giacomazzo JILA - University of Colorado 440 UCB Boulder, CO 80309 USA
Tel. : +1 303-492-5170 Fax : +1 303-492-5235 email : bruno.giacomazzo@jila.colorado.edu web: http://www.brunogiacomazzo.org
---------------------------------------------------------------------- There are only 10 types of people in the world: Those who understand binary, and those who don't ----------------------------------------------------------------------