[Wien] k-point parallel problem

Peter Blaha pblaha at zeus.theochem.tuwien.ac.at
Thu Jan 6 08:26:09 CET 2005


You have a problem with your queuing system. Aparently there is a time 
limit of 60 seconds, so small jobs (Si) run, longer don't. 
You need to check with your system manager.

> Dear Peter Blaha,
> 
>        This time we calculate the bulk of Si. which is much smaller.there is no problem in k-point parral. Only in the end the state flag is "E" when we use "qstat -a". We use 12 nodes and 24 CPUs, 74 k-points.
>        Then we test a larger system (16 atoms).
> 1,we run the program directly in the masternode. " run_lapw -i 1 -ec 0.0001", there is no problem.
> 2,Then we submit the program using pbs and run it in serial model.
>    qsub ./wienifc.pbs -l nodes=1:ppn=1 -q high -N sisiy4
>    the command line in wienifc.pbs is: "run_lapw -i 1 -ec 0.0001 >run.output"
>    I use the "top" comand look at the process. It gave an error message after the lapw1c run 1min.
>   "You should submit the program use pbs."
> 
> the dayfile is:
> 
> Calculating sisiy4 in /people/gong/lyhua/ADP/sisiy4
> on comp10 with PID 19621
> 
>     start       (Thu Jan  6 11:13:48 CST 2005) with lapw0 (1/20 to go)
> >   lapw0       (11:13:48) 31.840u 0.310s 0:32.58 98.6% 0+0k 0+0io 172pf+0w
> >   lapw1  -c   (11:14:20) Killed
> 38.870u 0.680s 0:39.85 99.2%    0+0k 0+0io 238pf+0w
> 
> >   stop error
> ~
> the lapw1.error is
> 
> Error in LAPW1
> 
> 3, then we test the parallel model.
>    qsub ./wienifc.pbs -l nodes=2:ppn=1 -q high -N sisiy4
>   the comand line in wienifc.pbs is:
>   "run_lapw -i 1 -ec 0.0001 -p > run.output"
>   as above it give an error message:
>   "You should submit the program use pbs"
> 
> ______________________________
> the case.dayfile is:
> 
> 
> Calculating sisiy4 in /people/gong/lyhua/ADP/sisiy4
> on comp10 with PID 16465
> 
>     start       (Thu Jan  6 10:38:00 CST 2005) with lapw0 (1/20 to go)
> >   lapw0 -p    (10:38:00) starting parallel lapw0 at Thu Jan  6 10:38:00 CST 2005
> --------
> running lapw0 in single mode
> 32.020u 0.350s 0:33.17 97.5%    0+0k 0+0io 1823pf+0w
> >   lapw1  -c -p        (10:38:33) starting parallel lapw1 at Thu Jan  6 10:38:33 CST 2005
> ->  starting parallel LAPW1 jobs at Thu Jan  6 10:38:33 CST 2005
> running LAPW1 in parallel mode (using .machines)
> 2 number_of_parallel_jobs
> **  LAPW1 crashed!
> 0.150u 0.220s 1:28.34 0.4%      0+0k 0+0io 21900pf+0w
> 
> >   stop error
> _______________________________________
> 
> the lapw1.error is:
> 
> 
>  **  Error in Parallel LAPW1
> **  LAPW1 STOPPED at Thu Jan 6 10:40:01 CST 2005
> **  check ERROR FILES!
> Error in LAPW1
> Error in LAPW1              
> _______________________________________
>  
> the end of case.output1 is:
> 
>   1.3497807    1.3891706    1.3891808    1.3933397    1.3935096
>       1.4114150    1.4118608    1.4128885    1.4133827
>             0 EIGENVALUES BELOW THE ENERGY   -7.00000
>        ********************************************************
> 
>        NUMBER OF K-POINTS:           1
>    ===> TOTAL CPU       TIME:    204.3 (INIT =      0.8 + K-POINTS =    203.6)
>    > SUM OF WALL CLOCK TIMES:    206.6 (INIT =      1.0 + K-POINTS =    205.5)
>       Maximum WALL clock time:    207.062600851059
>       Maximum CPU time:           204.480000000000 
> ________________________________________
>  
> the case.scf1 is empty.    
>         	
> 
> 
> 
> >First: please use a number of processors which is compatible with the
> >number of k-points you have (check case.klist). What I mean is: I suppose
> >you have 18 k-points, so reasonable machine-numbers are 18 (each processor
> >does 1 k-point), 9 (each does 2), 6, 3 or 2.
> >Of course you can use 16, but the program will be not faster (in fact even 
> >slower since summation takes longer) than with 9 processors.
> >
> >Second: Your message is not quite clear: it failed when run the lapw1para...
> >        There is no problem when run lapw1 in parral.   ???
> >
> >Third: In your script please change (sorry, it was incorrect on the faq page
> >       but it should not be responsible for the problems) 
> >echo 'extrafine' >>.machines         to
> >echo 'extrafine:1' >>.machines
> >                       
> >Fourth: Could be a timelimit the cause for these problems ? Increase the
> >cpu-time-limit of the pbs job. 
> >In addition your pbs job should produce output and error files and they may 
> >contain further information.
> >
> >
> >
> >                                      P.Blaha
> >--------------------------------------------------------------------------
> >Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
> >Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
> >Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
> >--------------------------------------------------------------------------
> >
> >_______________________________________________
> >Wien mailing list
> >Wien at zeus.theochem.tuwien.ac.at
> >http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> 
> = = = = = = = = = = = = = = = = = = = =
> 			
> 
>         ?
> ?!
>  
> 				 
>         liyh
>         lyhua at fudan.edu.cn
>           2005-01-06
> 
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> 


                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------




More information about the Wien mailing list