Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] calling sendi earlier in the PML
From: Eugene Loh (Eugene.Loh_at_[hidden])
Date: 2009-03-04 20:26:05


Brian W. Barrett wrote:

> How about removing the MCA parameter from my earlier proposal and just
> having r2 filter out the sendi calls if there are multiple BTLs with
> heterogeneous BTLs (ie, some with sendi and some without) to the same
> peer. That way, the early sendi will be bypassed in that case. But
> for the cases of BTLs that support sendi in common usage scenarios
> (homogeneous nics), we'll get the optimization? Does that offend you
> George? :)

I think I'm just going to punt. The PML strikes me as very complicated
and in a certain sense brittle. You talked in San Jose about putting
the PML on a diet. Great. Go for it. For a newbie like me, it's a
labyrinth.

Here's another example. Even if you only go for the cases where
everyone has (or does not have) a sendi function, they may have
different eager limits. (Though, somehow there is a well-defined list
of "eager" BTLs that does not depend on message length. Ain't life
interesting!) So, you run into the same problem of not preserving
today's BTL-looping order. If you want to preserve the current behavior
-- looping over BTLs and trying all your tricks for each one
(sendi/send, eager/long, etc.) before moving on to the next BTL --
you're back to diving deep into the PML code, at which point the send
request has been initialized and you've eaten up a lot of instructions.

I don't think I have sufficient expertise, mandate, or remaining energy
to be effective here.