Hi Joerg,
I recently started using Gadget2 and had the same trouble. The code
parameter which will remedy the problem is TreeAllocFac, which controls
how many tree nodes are allocated per processor. It is set at 0.8 as a
standard value, which will work for large numbers of particles per
processor. I first noticed this problem while I was testing CPU scalings
on a couple super computers and running somewhat small runs on large
numbers of processors (similar to the runs you quoted below actually).
Volker suggested increasing Treeallocfac from 0.8 to 1.2, or 1.5 to be
safe and that it shouldn't affect the performance of the code.
I hope this helps.
Aaron
_______________________________________________________________
Aaron D. Ludlow
Department of Physics and Astronomy Tel: +1-250-721-8819
University of Victoria Fax: +1-250-721-7715
3800 Finnerty Road email: ludlow_at_uvic.ca
Victoria, BC V8P 5C2 CANADA
_______________________________________________________________
On Tue, 11 Oct 2005, Joerg Colberg wrote:
>
> Hi guys,
>
> I've just started working with Gadget for the very first
> time, using a supercomputer at the Pittsburgh Supercomputer
> Center, and I'm having some initial and probably utterly
> trivial problems. I'm assuming I didn't set all the parameters
> right, but I'm afraid I don't know which one(s) cause(s) the
> problem.
> I'm trying to run a 2*128^3 gas run on 64 procs, and right
> at the start of the code the whole thing stops telling me:
>
> "domain decomposition...
> NTopleaves= 54272
> work-load balance=1.00932 memory-balance=1.00932
> exchange of 0004129449 particles
> task 59: maximum number 68812 of tree-nodes reached.
> for particle 24990"
>
> (there appears to be one such error for each proc)
>
> What did I do wrong?
>
> Best,
> Joerg
> ___________________________________________________________
> Dr. Joerg M. Colberg
> Carnegie Mellon University
> Department of Physics and Astronomy
> 5000 Forbes Avenue
> Pittsburgh PA 15213
> USA
>
> +1.412.268.8351
>
> http://lahmu.phyast.pitt.edu/~colberg
> http://www.jmcolberg.com/weblog
>
>
>
>
> -----------------------------------------------------------
>
> If you wish to unsubscribe from this mailing, send mail to
> minimalist_at_MPA-Garching.MPG.de with a subject of: unsubscribe gadget-list
> A web-archive of this mailing list is available here:
> http://www.mpa-garching.mpg.de/gadget/gadget-list
>
Received on 2005-10-11 17:13:20