Hi Hui. About the "special (or interactive) mode" would be great! In relation to your frontend-backend suggestion that you propose, having a "proxy" process in that way might be an important code restructuring which we just can't do at the moment. Thank you very much anyway. Best regards, Martín <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail> Virus-free. www.avast.com <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail> <#DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2> On Fri, Sep 25, 2020 at 12:58 PM Zhou, Hui <[email protected]> wrote:
Hi Martin,
I see. Yeah, MPICH (or the project manager hydra) will not work with interactive programs nicely. It might be a potential interest to add a special mode that project manager do not try to take over the input/output. For now, is it possible for your ncurse application to spawn a separate process (via `mpirun -n 1 …`) that can be used to spawn additional processes? I am thinking that your interactive part can be programmed as purely a frontend to your actual MPI processes.
-- Hui Zhou
*From: *Martín Morales <[email protected]> *Date: *Friday, September 25, 2020 at 9:09 AM *To: *"[email protected]" <[email protected]>, "Zhou, Hui" <[email protected]> *Subject: *Re: [mpich-discuss] Spawns without mpirun
Hi again Hui. My apologies, I overlooked your singleton analogie with mpirun (*mpirun -n 1 ./prog...* ). Yes, in fact, it was one our first tries. These program we use is interactive and requires the *ncurses *library for that. I read some time ago a known problem with Open MPI implementation and *ncurses*. Unfortunately with MPICH we've experienced the same issue.
Best regards,
Martín
On Wed, Sep 23, 2020 at 9:28 PM Martín Morales < [email protected]> wrote:
Hi Hui, thank you for your reply. Ok I'll create that issue then. We've a quite large PVM application. It has processes handling in a singleton fashion that it allows powerful functionality. We needed to port the PVM code to MPI in this exact way to preserve that functionality. We've done this already but with the Open MPI implementation. However we've found some inconsistencies in it and that's why I'm querying about this feature in MPICH.
Best regards
Martín
On Wed, Sep 23, 2020 at 8:31 PM Zhou, Hui <[email protected]> wrote:
It’s possible but not prioritized. Could you open an issue on github? When we have enough users requesting the feature (or when we have persistent user requesting it), the priority may be escalated. 😊 Meanwhile, could you describe the scenario that you have to use singleton init (vs. `mpirun -n 1 prog …`)?
-- Hui Zhou
*From: *Martín Morales via discuss <[email protected]> *Reply-To: *"[email protected]" <[email protected]> *Date: *Wednesday, September 23, 2020 at 3:49 PM *To: *"[email protected]" <[email protected]> *Cc: *Martín Morales <[email protected]> *Subject: *[mpich-discuss] Spawns without mpirun
Hi all!
I asked here some time ago about dynamically spawn processes from a singleton but the answer was that unfortunately, there was a problem with that and just It was not possible. I wonder now if this functionality is available.
Best regards
Martín
Virus-free. www.avast.com <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>