[179769] in North American Network Operators' Group
Re: link avoidance
daemon@ATHENA.MIT.EDU (Scott Whyte)
Wed May 6 21:08:58 2015
X-Original-To: nanog@nanog.org
Date: Wed, 06 May 2015 18:05:06 -0700
From: Scott Whyte <swhyte@gmail.com>
To: Randy Bush <randy@psg.com>,
North American Network Operators' Group <nanog@nanog.org>
In-Reply-To: <m2lhh18gdi.wl%randy@psg.com>
Errors-To: nanog-bounces@nanog.org
On 5/6/15 15:56, Randy Bush wrote:
> a fellow researcher wants
>
> > to make the case that in some scenarios it is very important for a
> > network operator to be able to specify that traffic should *not*
> > traverse a certain switch/link/group of switches/group of links
> > (that's true right?). Could you give some examples? Perhaps point
> > me to relevant references?
>
> if so, why? security? congestion? other? but is it common? and, if
> so, how do you do it?
My experience has been with MPLS overlays.
Availability: During maintenance windows, moving high-value traffic away
from potential outages while keeping the tunnels full of BE; manually
manipulating MPLS tunnel affinities (though this could be automated
fairly easily).
Congestion: Whenever traffic load spikes past a threshold;
diffserv-aware TE to prevent certain classes of traffic from routing
over links with limited bandwidth, handled automatically via auto-bw.
Preventing non-optimal tunnel paths. No transoceanic trombones,
please; MPLS link affinities designed into the network.
-Scott
