[e2e] IP options inserted in transit

[RJ Atkinson]

On Wednesday, Aug 20, 2003, at 04:21 America/Montreal, Michael Welzl 
wrote:
> In general, option processing _DOES_ make quite a difference
> for a large number of routers that are installed on the
> Internet of today.

Sure.  This is particularly true for the large installed base
of cisco products.  It is less true or untrue for routers made
by certain other vendors.  Some recent cisco products (e.g. 124xx)
might do better at forwarding IP packets with options.  I'm not
sure either way about that box since I haven't tested a cisco 124xx
box myself.

> Over the last three weeks or so, we did some pings (with
> and without a NOP option) with several thousands of hosts
> across a large range of path lengths, the goal being to
> figure out what "quite a difference" exactly means. What
> is missing is a statistical analysis - a student is working
> on it. We also have data from approx. 2 years ago, so we
> will also try to find out if the situation has changed.

A large percentage of the deployed routers forward IP packets
that contain any IP options through the "slow path".  In those
routers, only IP packets that contain no IP options are forwarded
through the "fast path".  This has been true for many years
and is partly an artifact of software-based forwarding.

I had thought this was widely known, but apparently not outside
of the NANOG/APRICOT/RIPE operational crowd...

Ran Atkinson
rja at extremenetworks.com