Qwalk with mpi

[Stefano Di Sabatino]

Dear All,

I am trying to compile the parallel version of Qwalk.

I used

 make PLATFORM=Linux-mpi

make install PLATFORM=Linux-mpi 

the file make/Linux-mpi.mk has not been modified.

The code seems to compile and I get qwalk in bin  and qwalk-Linux-mpi in src.

When I try to run the code with mpirun -n N ... it seem that the code is simply running N times in serial. And since the N runs are not synchronized the log file is messy and gosling is not able to read the data to make the averages (I get a segmentation fault).

Could anyone help me solving this problem?

Best regards,

Stefano

[Lucas Wagner]

Hi Stefano,

Thanks for the message! These days, we are slowly transitioning to using pyqmc: https://github.com/wagnergroup/pyqmc for new problems. It's quite a bit more convenient than QWalk and we are actively working on it. 

Your problem sounds like it could be due to a mismatch of your MPI libraries or something similar. I would try compiling a simple example MPI program using the same setting and see if that works.

Cheers,

Lucas

--
You received this message because you are subscribed to the Google Groups "QWalk user's group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qwalk-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/qwalk-users/d4037b60-c18e-4808-a20c-35cb17054758n%40googlegroups.com.

[Stefano Di Sabatino]

Dear Luca,

thanks for the reply. I compiled again the code using mpicxx, openmpi and openblas.

Now if I run the code using mpirun I have this output (I am using VMC)

cutoff 9.79796
 total centers 109
Ewald sum using 706 reciprocal points
Could not open si_bulk_0.hf.config. Generating configurations from scratch

I do not see any indication that the code is running in parallel. I am running the code on a single node with multiple cores.

And if I use more cores I do not see a reduction of the running time. The only change that I see is that the code is automatically setting nconfig equal to the number of cores I am using.

Does this mean that the code is actually running in parallel?

I also get this warning message

--------------------------------------------------------------------------
[[27043,1],5]: A high-performance Open MPI point-to-point messaging module
was unable to find any relevant network interfaces:

Module: OpenFabrics (openib)
  Host: compute-7-3

Another transport will be used instead, although this may result in
lower performance.

NOTE: You can disable this warning by setting the MCA parameter
btl_base_warn_component_unused to 0.
--------------------------------------------------------------------------

Best

Stefano

You received this message because you are subscribed to a topic in the Google Groups "QWalk user's group" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/qwalk-users/9JEJ7jkSrY0/unsubscribe.
To unsubscribe from this group and all its topics, send an email to qwalk-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/qwalk-users/CA%2B%3Dghi9oo2zUD-snV%2BUccLHWkts_%2B16LYUEBw9wqkwB-eRQu1g%40mail.gmail.com.

[Lucas Wagner]

Hi Stefano,

That looks good to me. The run time will not decrease because Monte Carlo paralellizes using replicas. So instead of reducing the run time, you will see your stochastic error decrease for a given number of steps*blocks.

Let me note again that qwalk hasn’t been updated or developed for a few years, and we are moving to pyqmc, which I think is much better!

Lucas

On Feb 24, 2021, at 3:29 AM, Stefano Di Sabatino <stef.di....@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/qwalk-users/CAMQp8q9rAB2%2Bno%3D3_Q8Bbq%2BczOZFkKpmerjAB1jiHD_f_0HA%3DQ%40mail.gmail.com.

[Stefano Di Sabatino]

thanks a lot!

I am interested mainly in the possibility of calculating the EKT energies of solids. 

In pyqmc I have not seen any reference to the EKT so I think at the moment Qwalk is still the best option for me.

Thanks again

Best

Stefano

To view this discussion on the web visit https://groups.google.com/d/msgid/qwalk-users/D0099CBC-8C1B-4101-9D9E-76E153BF9AC8%40gmail.com.