- Mail actions: [ respond to this message ] [ mail a new topic ]
- Contemporary messages sorted: [ by date ] [ by thread ] [ by subject ] [ by author ] [ by messages with attachments ]

From: Volker Springel <volker_at_MPA-Garching.MPG.DE>

Date: Mon, 08 May 2006 17:13:12 +0200

Brad Krane wrote:

*> Sorry for the post spamming, but I made a mistake with my last reply I
*

*> am actually running the cluster sample. I have also just discovered
*

*> that removing the double precision options from the make file without
*

*> increasing the PMGRID value has seemed to fix the problem (at least
*

*> for now) as I have run the same cluster simulation past the point
*

*> where it would fail at step 3 with the double precision options up to
*

*> step 112 currently without them. Is it possible that there is
*

*> something wrong with the double precision code that fixes the particle
*

*> out of range problem?
*

I think the behavior has only been different because of different

effects of floating point rounding. When you used double precision, the

new PM grid region was determined more accurately, so that the "particle

on the edge" problem could happen more easily, while it is less likely

to happen with single-precision rounding. The real solution to this is

to use ENLARGEREGION > 1.0.

Volker

*> Thanks.
*

*>
*

*> On 5/8/06, Brad Krane <brad.krane_at_gmail.com> wrote:
*

*>> Sorry I forgot. My compiling options are:
*

*>>
*

*>> OPT += -DUNEQUALSOFTENINGS
*

*>> OPT += -DPEANOHILBERT
*

*>> OPT += -DWALLCLOCK
*

*>> OPT += -DPMGRID=128
*

*>> OPT += -DDOUBLEPRECISION
*

*>> OPT += -DDOUBLEPRECISION_FFTW
*

*>> OPT += -DSYNCHRONIZATION
*

*>>
*

*>> And I am running the galaxy sample.
*

*>>
*

*>> On 5/8/06, Brad Krane <brad.krane_at_gmail.com> wrote:
*

*>> > I have been coming up to the same problem with Gadget2 now for a few
*

*>> > days now with out any real success. It seems that if I run a
*

*>> > non-periodic simulation long enough that one or more of the particles
*

*>> > leaves the PM which is usually fine at first but eventually Gadget is
*

*>> > unable to recover and the program halts with a 68687 error. I know
*

*>> > that one way to fix this problem is to increase the PMGRID size on
*

*>> > compile, but this seems to lead to two problems:
*

*>> >
*

*>> > 1. An exponential increase in memory use.
*

*>> > 2. The particle will just leave the now larger mesh later in the
*

*>> > simulation anyway.
*

*>> >
*

*>> > Is there anyway that I might be able to just kill the offending
*

*>> > particle completely from the simulation and thereby prevent the crash?
*

*>> > Would this approach lead to problems? Is there a better way that I am
*

*>> > not thinking about? Or should I just get a better computer? ;)
*

*>> >
*

*>> > Thank you.
*

*>> >
*

*>> > Here below is a sample of the output if that helps clarify my situation.
*

*>> >
*

*>> > Begin Step 2, Time: 0.0437879, Redshift: 21.8373, Systemstep:
*

*>> > 0.00107381, Dloga: 0.0248285
*

*>> > domain decomposition...
*

*>> > NTopleaves= 176
*

*>> > work-load balance=1 memory-balance=1
*

*>> > domain decomposition done.
*

*>> > begin Peano-Hilbert order...
*

*>> > Peano-Hilbert done.
*

*>> > Start force computation...
*

*>> > Starting non-periodic PM calculation (grid=0).
*

*>> > Particle Id=142373 on task=0 with coordinates
*

*>> > (-68861.4|-1927.9|-1800.34) lies outside PM mesh.
*

*>> > Stopping
*

*>> > In total 1 particles were outside allowed range.
*

*>> >
*

*>> > Allowed region for isolated PM mesh (coarse):
*

*>> > (-68861.4|-68906.6|-68905.6) -> (68906.8|68861.6|68862.6)
*

*>> > ext=137768 totmeshsize=286880 meshsize=1120.63
*

*>> >
*

*>> > Starting non-periodic PM calculation (grid=0).
*

*>> > done PM.
*

*>> > Tree construction.
*

*>> > Tree construction done.
*

*>> > Begin tree force.
*

*>> > tree is done.
*

*>> > force computation done.
*

*>> > type=1 dmean=510.503 asmth=1400.78 minmass=1.10779 a=0.0437879
*

*>> > sqrt(<p^2>)=6.81455 dlogmax=0.17174
*

*>> > type=2 dmean=496.112 asmth=1400.78 minmass=1.01672 a=0.0437879
*

*>> > sqrt(<p^2>)=4.92943 dlogmax=0.230724
*

*>> > type=3 dmean=185.921 asmth=1400.78 minmass=0.0535116 a=0.0437879
*

*>> > sqrt(<p^2>)=5.61213 dlogmax=0.0759471
*

*>> > displacement time constraint: 0.03 (0.03)
*

*>> >
*

*>> > Begin Step 3, Time: 0.0448887, Redshift: 21.2773, Systemstep:
*

*>> > 0.0011008, Dloga: 0.0248285
*

*>> > domain decomposition...
*

*>> > NTopleaves= 176
*

*>> > work-load balance=1 memory-balance=1
*

*>> > domain decomposition done.
*

*>> > begin Peano-Hilbert order...
*

*>> > Peano-Hilbert done.
*

*>> > Start force computation...
*

*>> > Starting non-periodic PM calculation (grid=0).
*

*>> > Particle Id=142373 on task=0 with coordinates
*

*>> > (-68871.7|-1927.62|-1803.53) lies outside PM mesh.
*

*>> > Stopping
*

*>> > In total 1 particles were outside allowed range.
*

*>> >
*

*>> > Allowed region for isolated PM mesh (coarse):
*

*>> > (-68871.7|-68914.7|-68906.5) -> (68899.7|68856.7|68864.9)
*

*>> > ext=137771 totmeshsize=286887 meshsize=1120.65
*

*>> >
*

*>> > Starting non-periodic PM calculation (grid=0).
*

*>> > Particle Id=142312 on task=0 with coordinates
*

*>> > (68899.7|-1661.64|1693.9) lies out side PM mesh.
*

*>> > Stopping
*

*>> > In total 1 particles were outside allowed range.
*

*>> > task 0: endrun called with an error level of 68687
*

*>> >
*

*>> > --
*

*>> > Brad
*

*>> >
*

*>>
*

*>>
*

*>> --
*

*>> Brad
*

*>>
*

*>
*

*>
*

*> --
*

*> Brad
*

*>
*

*>
*

*>
*

*>
*

*> -----------------------------------------------------------
*

*>
*

*> 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 2006-05-08 17:13:12

Date: Mon, 08 May 2006 17:13:12 +0200

Brad Krane wrote:

I think the behavior has only been different because of different

effects of floating point rounding. When you used double precision, the

new PM grid region was determined more accurately, so that the "particle

on the edge" problem could happen more easily, while it is less likely

to happen with single-precision rounding. The real solution to this is

to use ENLARGEREGION > 1.0.

Volker

Received on 2006-05-08 17:13:12

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