[50615] in North American Network Operators' Group
Re: If you have nothing to hide
daemon@ATHENA.MIT.EDU (Eric Osborne)
Mon Aug 5 23:49:51 2002
Date: Mon, 5 Aug 2002 23:49:20 -0400
From: Eric Osborne <eosborne@cisco.com>
To: bdragon@gweep.net
Cc: Eric Osborne <eosborne@cisco.com>, nanog@merit.edu
In-Reply-To: <20020805224659.82061.qmail@sidehack.sat.gweep.net>
Errors-To: owner-nanog-outgoing@merit.edu
On Mon, Aug 05, 2002 at 06:46:59PM -0400, bdragon@gweep.net wrote:
> >
> >
> >
> > > Validation of routing policy to ensure others aren't abusing you (pointing
> > > default, for example). As for orders of magnitude, once an IP option is
> > > in a packet, the damage is essentially done, otherwise looking up the
> > > path to an address in the options is no more impactive than looking up the
> > > address in the original destination field.
> >
> > Well, no. Not really.
> > First off, following the 80/20 rule (or in this case 99.x/(100-99.x)
> > rule) says that hardware implementations which get optioned packets
> > punt them to software. This is at every hop.
> >
> > Second, the IP source route is a stack of IP addresses, which must be
> > modified at every hop. This implies not just software forwarding, but
> > also significantly more work than an IP lookup.
>
> As I said, once the option is in the packet, the damage is done.
> If the performance sucks for the person using the source-routing, who
> cares, assuming packets without IP options are forwarded without
> delay.
You care...the more work you do in SW, the less time your SW has to do
useful things like make sure the HW is talking to the control plane.
This is either an argument for more HW support for optioned packets or
less optioned packets on the network, depending on your perspective.
>
> If I'm not mistaken, most (if not all) vendors still punt the
> packets with source-routing options to software, even if they end
> up dropping the packet due to administrative decision.
Yeah, generally, although it could certainly depend not only on vendor
but on engine...:)
eric
>
> > eric
