Network Stack Locking

[Robert Watson]

On Tue, 25 May 2004, Luigi Rizzo wrote:

> On Tue, May 25, 2004 at 11:20:49AM -0700, Matthew Dillon wrote:
> ...
> > :Actually, this was the specific point I was making also :-).  The question
> > :I was asking was about the depth of the message queues between protocol
> > :stack layers in actual measurements -- are you observing substantial
> > :coallescing between layers as a result of the queues at this point?  I'm
> ...
> >     It should also be noted that since today's processors are so damn fast,
> >     actually maxing out the cpu before maxing out the PCI bus(es), at least
> >     on a standard workstation, is difficult.
> 
> that's where the soekris comes handy :)  (yes i understand that much of
> the issues here are related to SMP, but there lot of evidence that a
> non-smp 5.x kernel is significantly slower than the equivalent 4.x code,
> even on heavy network i/o loads, so there is certainly something to
> clean up there too). 

Happily, I'm well endowed with old and crufty hardware when it comes to
UP. That said, it's hard to find a nice middle ground with SMP -- even
relatively slow boxes can typically can max out 32-bit PCI, so you need to
go with 64-bit if you want to not have that be the bottleneck for gig-E.

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Senior Research Scientist, McAfee Research