Re: RE: Re: Gadget stalling without apparent reason

From: Nathan Goldbaum <nathan12343_at_gmail.com>
Date: Mon, 31 Oct 2016 20:38:21 -0500

On Mon, Oct 31, 2016 at 8:26 PM, Raymond Carlberg <
raymond.carlberg_at_utoronto.ca> wrote:

> I have set the minimum time step to a value below what the dynamic range
> should reasonably need. Indeed all is well for about ½ of the longest
> dynamical time. The time step does not shrink to zero, it just drops in one
> step.
>

I'd trace through the code to figure out exactly where this is happening. I
usually do this using printf's or GDB.

It sounds vaguely like integer or floating point overflow, but that's just
a vague guess on my part.


>
>
> *From:* isbug01_at_gmail.com [mailto:isbug01_at_gmail.com] *On Behalf Of *ISAAC
> SHLOSMAN
> *Sent:* Monday, October 31, 2016 9:09 PM
> *To:* Gadget General Discussion <gadget-list_at_MPA-Garching.MPG.DE>
> *Subject:* Re: [gadget-list] RE: Re: Gadget stalling without apparent
> reason
>
>
>
> Hi Raymond,
> This because your extreme dynamic range and a finite number of time bins
> results in dt going to 0. Find a way to decrease dynamic range
>
> Isaac
>
>
>
> On Oct 31, 2016 7:45 PM, "Raymond Carlberg" <raymond.carlberg_at_utoronto.ca>
> wrote:
>
> HI – I am running Gadget2 on a 32 node machine. Works well for more than
> 10,000 steps, then dt goes to zero with no complaints from the program so
> the run continues for some time with no progress. DONOTSTOP is not
> implemented. Softenings are good. TreeAllocFactor boosted, etc, but no go .
> The gravity calculation has extreme density ranges.
>
> --Ray
>
>
>
> Restarting shows a possible problem (Nf=0 below):
>
> Step= 9 t= 1.55115 dt= 0.000128937
>
> Nf= 0000064647 total-Nf= 0001622692 ex-frac= 21.9985 iter= 1
>
> work-load balance: 18.532 max=4.10968 avg=0.221762 PE0=1.98117
>
> particle-load balance: 4.77931
>
> max. nodes: 37131, filled: 0.0066013
>
> part/sec=4554.93 | 245.788 ia/part=5650.51 (0)
>
>
>
> Step= 10 t= 1.55115 dt= 0
>
> Nf= 0000000000 total-Nf= 0001622692 ex-frac= -nan iter= 0
>
> work-load balance: -nan max=0 avg=0 PE0=0
>
> particle-load balance: 4.77931
>
> max. nodes: 37131, filled: 0.0066013
>
> part/sec=0 | -nan ia/part=-nan (-nan)
>
>
>
> ^^^^^^^^^^^^^^^^^^^and the output at the time of the problem.
>
>
>
> Begin Step 13838, Time: 1.55121, Systemstep: 0.000152588
>
> domain decomposition...
>
> NTopleaves= 4124
>
> work-load balance=36.5349 memory-balance=4.45177
>
> exchange of 0000288372 particles
>
> domain decomposition done.
>
> begin Peano-Hilbert order...
>
> Peano-Hilbert done.
>
> Start force computation...
>
> Tree construction.
>
> Tree construction done.
>
> Begin tree force.
>
> tree is done.
>
> force computation done.
>
>
>
> Begin Step 13839, Time: 1.55121, Systemstep: 0
>
> Start force computation...
>
> Begin tree force.
>
> tree is done.
>
> force computation done.
>
>
>
>
>
>
>
> -----------------------------------------------------------
>
> 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
>
>
>
> -----------------------------------------------------------
>
> 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 2016-11-01 02:38:44

This archive was generated by hypermail 2.3.0 : 2022-09-01 14:03:43 CEST