[e2e] "PMTUD using options" draft

[Nicolas Christin]

On Fri Feb 13, 2004, Lloyd Wood <l.wood at eim.surrey.ac.uk> wrote:
> On Fri, 13 Feb 2004, Jon Crowcroft wrote:
> 
> > routers that send anything to a "central" processor are going to
> > have a backplane or switch hotspot/bottleneck problem as well as a
> > cpu problem and are out of the ark designs/dinosaurs for sure....
> 
> These days, laptops are multihomed - supporting multiple wired and
> wireless links. But everything goes to a central processor, and
> so laptops are out-of-the-ark dinosaurs as far as routing goes.
> 
> I do hope nobody is proposing routing research based around using
> laptops for e.g. adhoc mesh communication. Or peer-to-peer systems
> based on these obviously dinosaur general-purpose computers.

I suspect that you are likely to hit another bottleneck before IP
options processing (that's still the subject of this thread, isn't it?)
becomes a serious worry in the context(s) you are referring to, which
is entirely different from what Ran and Jon were talking about.

1 Gbps is a relatively small link speed for a router, by today's
standards, right? Well, try to saturate a 1 Gbps Ethernet card on your
laptop with traffic, without using jumbo frames, while running an
overlay protocol. Good luck.

As an example, a Xeon running FreeBSD can
currently send in the order of 400 Kpps (see
<http://unix.derkeiler.com/Mailing-Lists/FreeBSD/net/2003-12/0089.html>)
. With an average packet size of 1000 bytes, that's about 3 Gbps if I'm
not mistaken, and that's when the PC is (as a first-order approximation)
only busy sending traffic. Imagine it has to do some processing with
the packets received before re-sending them, as is the case in a
peer-to-peer network or even when you're using your PC as a PC-router...

And if you're going to run a heavy-weight "user" application (e.g.,
Mozilla) while using your PC as a router, I would contend that IP
options processing is the least of your worries.

Or did I misunderstand your point?
-- 
Nicolas