Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] Consequence of bind-to-core by default
From: Ashley Pittman (ashley_at_[hidden])
Date: 2013-12-19 09:15:14

On 19 Dec 2013, at 13:59, Jeff Squyres (jsquyres) <jsquyres_at_[hidden]> wrote:
> - if we oversubscribe, (possibly) warn about the performance loss of oversubscription, and don't bind
> - don't warn about lack of memory binding
> Thoughts?

+1, I hit this myself today. I typically run on a VM and oversubscribe the cores, until the last update this would work fine, but now I get two error messages when trying this. I can’t “modify” the binding options used because I don’t know what they are (i.e. I didn’t give any) and even when not over-subscribing there is a warning at startup that I neither understand nor can seemingly disable.

My thoughts would be:

Oversubscription is normally bad so by all means issue a warning and/or abort however make the message meaningful and offer the use a —allow-oversubscription flag.

Jobs running on VMs shouldn’t give warnings to the user.

Finally, the whitespace alignment of the message is a little odd, it looks like it’s supposed to be a table or two columns however the indentation is all over the place.