[ase-users] NEB parallel with Abinit parallel

[Ask Hjorth Larsen]

Dear both,

Am Mo., 19. Nov. 2018 um 17:39 Uhr schrieb Offermans Willem
<willem.offermans at vito.be>:
>
> Dear Ask, Ivan, and ASE-users,
>
> I’m not sure that I understand the response of Ask.
>
> I like to run both NEB and ABINIT in parallel. So, for example, I would reserve 85 nodes and run 5 images on 13 nodes (= number of k-points) separately. ASE coordinates the NEB and optimisation of the NEB, ABINIT is the calculator and runs for several images. Each ABINIT calculation runs in parallel and is distributed over nodes by individual k-point. That is the basic idea.
>
> If I understood correctly, this is not yet supported by ASE. However, Ask is willing to support us (me), if I take the initiative to develop this.
>
> What do I need to do? And how can Ask support me?

The ase.calculators.socketio.SocketIOCalculator is a calculator which
wraps a server.  This server currently supports a single connection.
The connection represents a remote calculator.  If/when we need it, we
can expand it so it runs with multiple connections and hence can
control multiple remote calculators.  That would be useful for a NEB
run.

Now if a client process runs an ase.calculators.socketio.SocketClient
on an Atoms object with Abinit(command='mpirun ....') attached as
calculator, then the client's subprocess will run its calculations in
parallel, and yet communicate with the server as a single entity.

If you try it out on something simple like a relaxation, it should be
more clear.  It will work locally on a desktop computer.

  https://wiki.fysik.dtu.dk/ase/ase/calculators/socketio/socketio.html

So the goal would be to set up a NEB run where each atoms object
contains a SocketIOCalculator (which preferably would run a single
server rather than a server for each image; this is where something
needs to be modified), and then different processes can connect to it
and run independently.  Right now the processes may end up waiting for
each other inefficiently or something like that, but one would need to
see those problems first in order to fix them :)

Best regards
Ask

>
>
>
> Met vriendelijke groeten,
> Mit freundlichen Grüßen,
>
>
> Willem Offermans
> Researcher Electrocatalysis SCT
> VITO NV | Boeretang 200 | 2400 Mol
> Phone:+32(0)14335263 Mobile:+32(0)492182073
>
> Willem.Offermans at Vito.be
>
>
> On 19 Nov 2018, at 16:39, Ask Hjorth Larsen via ase-users <ase-users at listserv.fysik.dtu.dk> wrote:
>
> Hi,
>
> Am Mo., 19. Nov. 2018 um 14:51 Uhr schrieb Kondov, Ivan (SCC) via
> ase-users <ase-users at listserv.fysik.dtu.dk>:
>
>
> Dear all,
>
>
>
> I think that the ASE/NEB and the Calculator can run both in parallel only if the Calculator is GPAW. This is because the NEB and the process spawned by the Calculator have to share an MPI communicator (I am not sure about any existing tricks with application/machinefiles and mpirun options). Thus parallel NEB will work with VASP or Abinit in serial mode only. Also serial NEB will work with the parallel versions of VASP, Abinit etc.
>
>
>
> In order to run a parallel NEB with parallel VASP, the NEB implementation in the VTST package can be used (http://theory.cm.utexas.edu/vasp/) and VASP has to be patched. For Abinit I do not know such a solution.
>
>
>
> A generic solution with ASE could be to split the calculations of the images in each BFGS iteration into several jobs and distribute them on the cluster. After all jobs are ready then their outputs are collected and passed to BFGS and the next image geometries are computed, distributed and so on until BFGS converges. Unfortunately I do not have an implementation yet but I would prefer (and recommend) to use a workflow system such as FireWorks for this.
>
>
> We can perhaps modify the new SocketIOCalculator to accept and control
> multiple connections.  That would make it possible to launch a server
> (on any computer) and then a job with, say, 10 nodes, and each node
> connects to the server in order to receive positions and then
> calculate forces.  I can help if any of you are willing to take the
> initiative.
>
> Best regards
> Ask
>
>
>
>
> Best regards,
>
> Ivan
>
>
>
>
> _______________________________________________
> ase-users mailing list
> ase-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/ase-users
>
>
> Indien u VITO Mol bezoekt, hou aub er dan rekening mee dat de hoofdingang voortaan enkel bereikbaar is vanuit de richting Dessel-Retie, niet vanuit richting Mol, zie vito.be/route.
> If you plan to visit VITO at Mol, then please note that the main entrance can only be reached coming from Dessel-Retie and no longer coming from Mol, see vito.be/en/contact/locations.
> VITO Disclaimer: http://www.vito.be/e-maildisclaimer