Re: [swanmodel-users] Swan fails to merge output
Brought to you by:
mzijlema
|
From: João A. <ja...@au...> - 2020-06-05 00:18:15
|
Hello Clement, I have no experience with COAWST. However, based on my knowledge on linux/unix systems, I would say that the ulimit parameter is on the OS-level. That is, by using ulimit, It is possible for you as an user to set this value and make the OS allow SWAN to open this number of files. However (i) I don't believe a user without administrator privilege is able to set such a high number of files to be open, (ii) even if you are able, SWAN will still remain with its own limitation, which is in the swaninit file. I'd advise you to ensure that when you pass the ulimit parameter you (i) check whether this value is being accepted by the system and if so (ii) whether submitting your job with this parameter it effectively changes your swaninit file. The number of open files would be approximately the sum of your boundary files, wind forcing files, bathymetry file and output files. However I'm not sure how this number would be defined when running SWAN on MPI or OMP, i.e., is the total number of open files calculated based on each thread/process or on a single swan run? In my case I simply increased the number to a value greater than the sum I suggested earlier (I was running SWAN with OMP using 24 cores). Hope this helps... On Thu, 4 Jun 2020 at 23:13, Clement Calvino <cle...@uc...> wrote: > Dear João, > > Thank you for your reply, the "highest file ref. number" value is already > very high (99999999). How can I guess the number of files generated I'll be > dealing with during my run though? The number of output and input files > obviously matter, but I can't guess the impact of the number of processors > and the number of time steps in my run (which seems to matter). > > I am using the SWAN version within COAWST, I have the print files (per > processor) but they usually don't show any error message, sometimes Error > files are generated by SWAN but none are generated when SWAN fails to merge > the output. I can sometimes find an error message in the standard output > (log) of the run generated by COAWST, but it is not consistent. > > Kind regards, > Clement > > On Wed, 3 Jun 2020 at 23:51, João Albuquerque <ja...@au...> > wrote: > >> Dear Clement, >> >> When I went through the "Too many open files" message _in a completely >> different setup_ I changed the "highest file ref. number" in my swaninit >> file. SWAN creates this file at its first run, so check the value you have >> and increase it to a number that is compatible with the amount of files >> you'll be dealing with during your runs. >> >> I see you are looking for error messages. Are you looking for them in >> SWAN's .prt and .erf files? >> >> Best, >> João >> >> On Thu, 4 Jun 2020 at 05:21, Clement Calvino < >> cle...@uc...> wrote: >> >>> Dear Users, >>> >>> I have tried recently to run long simulations of my SWAN stand-alone >>> model for a coastal application (4 months long with 180s time-step). But if >>> short runs are terminating without issue for some reasons when running >>> longer simulations SWAN fails to merge properly the output files. I am >>> using the SWAN version within COAWST, but running SWAN as a stand-alone >>> application. >>> >>> Very rarely I manage to get an error message about "Too many open >>> files", I have tried without success to increase the limit when submitting >>> my job (ulimit -n 999999). >>> At best I managed to get a merged output file but the files per >>> processor were not deleted, it happened when I asked for one single table >>> output. I have tried reducing the duration of the run to only two or one >>> months, reducing the number of processors, number of output files and >>> number of boundary input files but nothing solved the issue... >>> >>> What annoys me is that I don't know what parameters I should focus on to >>> reduce the number of files opened during the run, would anyone know which >>> subroutine could be responsible for opening massive amounts of files? >>> >>> Thank you in advance, >>> Clement >>> >>> -- >>> Clement Calvino >>> PhD Student >>> School of Mathematics and Statistics >>> University College Dublin >>> _______________________________________________ >>> swanmodel-users mailing list >>> swa...@li... >>> https://lists.sourceforge.net/lists/listinfo/swanmodel-users >>> >> >> >> -- >> João Claudio Albuquerque >> PhD. Candidate, School of Environment, The University of Auckland, >> Level 4, Room 449, Science Centre, 23 Symonds Street, Auckland, NZ. >> +64 021 02365705 >> _______________________________________________ >> swanmodel-users mailing list >> swa...@li... >> https://lists.sourceforge.net/lists/listinfo/swanmodel-users >> > > > -- > Clement Calvino > PhD Student > School of Mathematics and Statistics > University College Dublin > -- João Claudio Albuquerque PhD. Candidate, School of Environment, The University of Auckland, Level 4, Room 449, Science Centre, 23 Symonds Street, Auckland, NZ. +64 021 02365705 |