[SCM] Samba Shared Repository - branch master updated

[Jeremy Allison]

On Fri, Jan 06, 2012 at 05:41:55PM -0500, Ira Cooper wrote:
> On Fri, Jan 6, 2012 at 5:40 PM, Jeremy Allison <jra at samba.org> wrote:
> 
> > On Fri, Jan 06, 2012 at 10:29:27PM +0100, Volker Lendecke wrote:
> > > On Fri, Jan 06, 2012 at 09:59:13AM -0800, Jeremy Allison wrote:
> > > > > Hmmm. I don't think this is the right approach. Think 1 CPU,
> > > > > 16 disks in a fat raid0. You want all spindles busy I think.
> > > >
> > > > Ok, how do you suggest this is tuned ?
> > >
> > > Don't know. My gut feeling is that we will end up with a
> > > global credit algorithm pretty soon. I think the aio module
> > > is too low-level for this, we should just spawn threads as
> > > more aio requests come in. For smb1 I am not too worried
> > > that we get overflowed, but with smb2 the picture might
> > > change. But there we have the credit algorithm to slow down
> > > clients when we get overflowed. I have no clue yet how this
> > > might work, but I think that is the right place to tackle
> > > the problem of getting flooded.
> >
> > Ok, for now I've allowed a "aio num threads" module
> > parameter to override the num cores heuristic.
> >
> > Or should I just set this to the number of credits
> > allowed in SMB2 ?
> >
> >
> No.  I believe that number is over 1024.

Ok, I've removed the num_cpus code and just set the
default to 100 (essentially unlimited for SMB1 as
we're limited by the max mux) and let the SMB2 crediting
algorithm take care of SMB2.