Re: Tree memory parameters and PMGRID size

From: Jay Billings <>
Date: Thu, 11 Sep 2008 20:09:05 -0400

Dear Mark,

I might be able to answer your questions about the PMGRID. The value of
PMGRID is the size of the grid in each dimension. The PM routine in
Gadget is used to determine the long-range forces, so this grid spans
the entire volume.

About your first question: Have you tried decreasing the number of
processors? I have experienced the same "maximum number of nodes-reached
error" many times. I can easily reproduce this error with the 60,000
particle galaxy collision test case by running it on twenty processors.
However, if I decrease the number of processors to four or so, then the
problem runs in about fifteen to thirty minutes without error. I
normally run 100,000 particles on fifteen processors and it takes about
an hour.

Hope this helps!


Mark Baumann wrote:
> Hello Gadget users,
> I would appreciate some advice on how to properly set the memory
> allocation parameters in the parameter file. I receive errors along
> the lines of "No domain decomposition that stays within memory bounds
> is possible" or "maximum number of top-level nodes reached".
> I've read through the User's Guide and the mailing list archive. I see
> that the relevant parameters are probably PartAllocFactor and
> TreeAllocFactor. Increasing these doesn't always seem to solve the
> problem. Values for TreeAllocFactor ranging from 0.7 all the way to
> 1000 don't seem to fix the problem, whether I try a large run (several
> thousand particles per processor) or a very small run (about 1
> particle per processor).
> In general...
> If you have a large # of particles per processor, what are recommended
> values for PartAllocFactor and TreeAllocFactor?
> On the opposite extreme, if you have a small # of particles per
> processor, how should I set the allocation factors?
> On a different note, about the size of the PM grid that is set by the
> "PMGRID=###" compiler option... is that the size of the PM grid over
> the entire computational volume, or is that the size of the PM grid
> per processor? Also, is it the size of the PM grid in each dimension,
> or is it the total number of grid cells? For the sake of a concrete
> example, what would you recommended for a grid size for a system of
> 100,000 particles on 512 processors?
> Thank you for your help!
> Mark
> -----------------------------------------------------------
> If you wish to unsubscribe from this mailing, send mail to
> with a subject of: unsubscribe gadget-list
> A web-archive of this mailing list is available here:
Received on 2008-09-12 02:41:12

This archive was generated by hypermail 2.3.0 : 2023-01-10 10:01:30 CET