My problem with dealii parallel scaling

[Mohammad Fazli]

Hi everyone,

I am using dealii and I faced a big problem with parallel scaling on ubuntu 18.04.

For instance, I have compiled step-17 with different core numbers on my machine:

Memory: 31.2 GB,

Processor: Intel® Core™ i7-8700 CPU @ 3.20GHz × 12,

which shows me:

time mpirun -np 1 ./step-17

real 0m17.701s

time mpirun -np 2 ./step-17

real 0m16.507s

time mpirun -np 4 ./step-17

real 0m33.934s

I have tried to install dealii several times but all the time the same issue happens. Someone told me the problem can be with p4est. But I don't know what I should do to resolve this issue.

Thank you for your help in advance.

Mohammad

[Wolfgang Bangerth]

On 2/9/22 01:49, Mohammad Fazli wrote:
> I am using dealii and I faced a big problem with parallel scaling on
> ubuntu 18.04.
> For instance, I have compiled step-17 with different core numbers on my
> machine:
> Memory: 31.2 GB,
> Processor: Intel® Core™ i7-8700 CPU @ 3.20GHz × 12,
> which shows me:
>

> *time mpirun -np 1 ./step-17*
> *real 0m17.701s*
> *time mpirun -np 2 ./step-17*
> *real 0m16.507s*
> *time mpirun -np 4 ./step-17*
> *real 0m33.934s*
> *
> *

> I have tried to install dealii several times but all the time the same
> issue happens. Someone told me the problem can be with p4est. But I
> don't know what I should do to resolve this issue.

Mohammad: step-17 does not actually use p4est, so that is not the cause.
Your results match what I see on my machine reasonably closely -- this
is what I get here:

1 processors: 0m25.833s
2 processors: 0m20.885s
4 processors: 0m14.838s
8 processors: 0m15.463s
16 processors: 0m15.858s
32 processors: 0m17.176s

In other words, there is not very much speedup. There can be many
reasons for that, among which are:

* You have only a single CPU. Using multiple MPI jobs will eventually
run into memory bandwidth limitations between the CPU and main memory.

* The problem is quite small (on the finest level, step-17 has only
about 40,000 unknowns). Breaking it into even smaller pieces does not
leave very much work for each MPI process, but requires substantial
communication.

* The program is intended to demonstrate some basic functionality, but
it is not tuned for efficiency, and certainly not for such small problems.

So there may not actually be anything wrong with what you do. If you
wanted to have a more realistic test, make sure that you increase the
initial number of refinement steps so that you have several 100,000
unknowns, and then try again with that.

Best
W.

--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/

[Mohammad Fazli]

Thanks, Dear Wolfgang for your reply,

I always have the problem with dealii installation as well:

I use candi for the installation as I have not found a straightforward installation process on dealii documentation.

It might be possible for my case the problem comes from my installation. Therefore, how do you usually install dealli on Ubuntu along with its dependencies (like P4est, Trilinos, METIS, ...)?

I am struggling with this issue for almost 3 weeks without any progress.

[Wolfgang Bangerth]

On 2/9/22 17:12, Mohammad Fazli wrote:
> It might be possible for my case the problem comes from my installation.
> Therefore, how do you usually install dealli on Ubuntu along with its
> dependencies (like P4est, Trilinos, METIS, ...)?

I personally install everything by hand, mainly because I've always done
it that way. But that's a stupid approach, and using candi is exactly
what you should be doing.

As for the error you show, you need to scroll up to see what the error
actually is. What you show only says "an error has happened at some time
before", but you need to find what that error actually is.

[Luca Heltai]

Another option you have on Ubuntu is to install deal.ii with apt-get. If you follow the instructions of our web site, it is straight forward and much faster.

Luca

> Il giorno 10 feb 2022, alle ore 03:20, Wolfgang Bangerth <bang...@colostate.edu> ha scritto:

> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en
> --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to dealii+un...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/dealii/8fcafe24-1426-b00a-9039-d3731bdbfc14%40colostate.edu.

[Mohammad Fazli]

Yes. I have tried that.

But since I need dealii for Lethe, I require to install its dependencies as well. To this end, it was really baffling to me how to specify the other dependencies directory path.