[Users] How to choose the correct parameters “ppn, max-num-threads, num-threads, nodes, --cores, --num-threads and –procs" on my computer?

[Roland Haas]

Hello Kuark,

Typically we see no benefit using hyperthreads due the nature of the
workload of our scientific codes.

So I will only describe things without hyperthreads (which mostly just
add a complications).

With that in mind simfactory's './simfactory/bin/sim setup-silent'
should do a good job at setting up a default machine description.

When you submit you should set ppn-used to the number of compute cores
you will use, one per OS thread. So with 16 threads you would use
ppn-used to 16 as well. 

If you are looking to get most speed on a typical workstation (single
node) then most likely you fastest way is actually to use 1 thread per
MPI rank and as many MPI ranks as you have physical cores.

So something like:

--cores 6 --num-threads 1 --ppn-used 6

in your case.

Yours,
Roland

> Hello everyone!
> 
> 
> 
> 
> I’m new to the ET. I have the ET installed on my desktop. It has one
> Intel(R) Core(TM) i5-10500 CPU with 6 physical cores and two threads
> per single physical core (but according to the Ubuntu’s conventions,
> the computer has 12 processors in total, as shown in the attachment
> of this email) and it also has Ubuntu 20.4. So I wonder whether you
> could please tell me how to set the correct parallel parameters for
> ET simulations, such as the options of the submit command
> ./simfactory/bin/sim (--cores, --num-threads and –procs) and the
> initial values of the file <username>.ini (pn, max-num-threads,
> num-threads, nodes).      
> 
> 
>  
> 
> 
> Up till now, I have tried many possible combinations without success.
> It usually returns messages like “Warning: Too many threads per
> process specified: specified num-threads=16 (ppn-used is 1)”,
> “Warning: Number of used cores per node, number of SMT threads, and
> number of threads per process are inconsistent: ppn-used=1,
> num-smt=1, num-threads=16_threads (ppn-used*num-smt must be an
> integer multiple of num-threads)” and “Warning: Total number of
> threads and number of threads per process are inconsistent: procs=1,
> num-threads=16 (procs*num-smt must be an integer multiple of
> num-threads)” and doesn't start running.    
> 
> 
>  
> 
> 
> I know parallelization plays a very important role in massive
> numerical computations like the ET. However, I’m very confused by
> many of the concepts it uses. I don’t find a documentation for these
> parameters after browsing websites (like
> https://urldefense.com/v3/__https://bitbucket.org/simfactory/simfactory2/src/master/doc/userguide/__;!!DZ3fjg!-V4c4apw4yKOJIfbQ1lrgx6hPKu9LiVqj0ILFwv_OEVnXTWKSbPZC8pb0g9powxNY4YgaE9EJD1NMwfPeCp1DnSGG_eq$
> ,
> https://urldefense.com/v3/__https://simfactory.org/info/documentation/__;!!DZ3fjg!-V4c4apw4yKOJIfbQ1lrgx6hPKu9LiVqj0ILFwv_OEVnXTWKSbPZC8pb0g9powxNY4YgaE9EJD1NMwfPeCp1DukO6jh5$
> ,
> https://urldefense.com/v3/__http://einsteintoolkit.org/usersguide/UsersGuide.html__;!!DZ3fjg!-V4c4apw4yKOJIfbQ1lrgx6hPKu9LiVqj0ILFwv_OEVnXTWKSbPZC8pb0g9powxNY4YgaE9EJD1NMwfPeCp1DhQA-A7m$
>  or even in the source code in the directory
> “./simfactory/lib/”).  There are merely some hints in the
> jupyter tutorial “CactusTutorial.ipynb”.
> 
> 
> 
> 
> 
> 
> 
> Kuarks


-- 
My email is as private as my paper mail. I therefore support encrypting
and signing email messages. Get my PGP key from http://pgp.mit.edu .
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <http://lists.einsteintoolkit.org/pipermail/users/attachments/20230713/f417a026/attachment-0001.sig>