[mpich-devel] about MPI hydra process management

[Balaji, Pavan]

We are not very good at names.  :-)

"persist" is for a "persistent daemon", which makes no sense, since sshd is a persistent too.

"bss" stands for "bootstrap server", and the name conflict is unfortunate.

  -- Pavan

From: unix haha
Reply-To: "devel at mpich.org<mailto:devel at mpich.org>"
Date: Tuesday, June 30, 2015 at 9:35 PM
To: "devel at mpich.org<mailto:devel at mpich.org>"
Subject: Re: [mpich-devel] about MPI hydra process management

Thanks, Balaji.

I took a quick look on persist bootstrap. It seems what I want and I will try it.

one thing puzzles me is what's "persist" means here for hydra. what's means for hydra_bss_persist in Makefile.mk under hydra/tools/bootstrap/? the bss is the .bss section in executable file?

Thanks!

On Wed, Jul 1, 2015 at 7:29 AM, Balaji, Pavan <balaji at anl.gov<mailto:balaji at anl.gov>> wrote:

Hydra supports a bunch of daemon processes, including ssh.  There is a daemon process included with MPICH called "persist", which you could use for this purpose, but it's not really maintained.  You could try it.

  -- Pavan

From: unix haha
Reply-To: "devel at mpich.org<mailto:devel at mpich.org>"
Date: Tuesday, June 30, 2015 at 6:18 PM
To: "devel at mpich.org<mailto:devel at mpich.org>"
Subject: [mpich-devel] about MPI hydra process management

Hi, all:

     I'm working on port the MPICH2 3.1.4 to an embedded environment. I met some problems for hydra process management, Could anyone help me?

     1. For the old MPD, before the mpirun, there's a daemon process running on all computer nodes. But I don't see what's hydra's daemon process in the current MPI document. Could you give me some hints?

      2. As we run the MPICH2 in a resource constraint environment, the SSH or the other linux-based launcher may not be used. Does it exist any simple launcher running on the computer nodes before the mpiexec?

      3. What's my understanding for hydra is: once the Launcher process running on all computer nodes, I can use the mpiexec on the control node to start the parallel jobs? If so, the problem is: Could I start the Launcher process manually if there's no linux-based launcher?

      Thanks!

_______________________________________________
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/devel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpich.org/pipermail/devel/attachments/20150701/43ea445a/attachment.html>