[Pipet Devel] BL PL design documentation

[Jean-Marc Valin]

Jarl van Katwijk a écrit :
> 
> Jean-Marc Valin wrote:
> 
> > > I don't think this is correct. Jarl's document seems to make an explicit
> > > point that there is only one BL (Parent) per Piper program running. So
> > > the DL will only communicate with one BL. The BL itself will be
> > > responsible for splitting processes up using the whole Mother to Baby
> > > thing. At least, this is my understanding.
> >
> > There will be one grandma-BL per Piper program running, however there will be
> > many mother-BL, one per machine, and many baby-BL per machine. Is that what you
> > meant?
> >
> 
> Arf, JM, maybe I'm still not clear about this. The functionality of scheduling you
> want the
> GrnatMother BL to do, will actually be done by what you call the Mother BL. So
> there's
> no point having the GrantMother-BL neither the Mother-BL names, that's why I call
> both just plain BL. I mean, there's no point naming features, or functionality.
> 
> Am I clear now, and, if you disagree, maybe we should have one more irc session
> about this ;)

Well, I would like to think that "all the BL's are born equal". To me, the BL on
the local machine shouldn't have more to do than any other BL. If you don't like
the name grandma-BL, we could call it "DL2BL interface". It would be a library
used by the DL. I'd like the scheduling to be done "in DL space". That's what I
meant by grandma-BL: one (small) part of the BL code that would be unique for a
piper program (while you have many parent-BL, one per machine). In my mind, we
would not even need a (parent-)BL on the local machine if we don't want to do
any processing there.

	Jean-Marc

-- 
Jean-Marc Valin
Universite de Sherbrooke - Genie Electrique
valj01 at gel.usherb.ca