[Wien] Problem with k-parallel in version 24.1?

[Peter Blaha]

> In case.dayfile, it may show something like this at the cycle where it 
> crashes:
> 
>        bb4u14c1 10.023u 17.606s 28.77 96.02% 0+0k 0+0io 0pf+0w

A timing statement like this tells you immediately that the job is not 
running well.
Your program (lapw2) needs 17 sec cpu time, but there is an additional 
10 second !!! system time (probably waiting for I/O).
The system time should always be much smaller.

....
> 
>        bb4u20c1 8.505u 1.509s 10.77 92.90% 0+0k 0+0io 0pf+0w
> 
>     Summary of lapw2para:
> 
>     bb4u14c1      user=452.054    wallclock=84644
> 
>     bb4u15c1      user=8.589      wallclock=712.33
> 
> **  LAPW2 crashed!
> 
> 3.440u 20.832s 1:12.41 33.5%    0+0k 0+67272io 12pf+0w

And this is even worse: Your (failed) job needed 72 seconds, although 
the cpu time of one step is only 17 seconds.

Always check the timing in the dayfile.   MORE parallelization 
(k-points, omp, mpi-cores) does NOT NECESSARILY mean, that it will run 
faster, but it can even dramatically SLOW DOWN.

PS: Often, the speed is also limited by memory access, so it could be 
that on a 64 core machine you should actually use only 32 .... But all 
these things are really hardware specific and you MUST do your personal 
timing tests to find out the best setup.

> 
> error: command   /home/yz155/WIEN2k_24.1/lapw2cpara -dn -c dnlapw2.def   
> failed
> 
> 
>  >   stop error
> 
> 
> As one can see for example, some cores on node bb4u14c1 did not run its 
> lapw2 -dn job, therefore no corresponding scf2dn file produced. The 
> dnlapw2 def files were produced though.
> 
> 
> 
> Best regards
> 
> Yichen
> 
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at:  http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html

-- 
-----------------------------------------------------------------------
Peter Blaha,  Inst. f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-158801165300
Email: peter.blaha at tuwien.ac.at
WWW:   http://www.imc.tuwien.ac.at      WIEN2k: http://www.wien2k.at
-------------------------------------------------------------------------