[97591] in North American Network Operators' Group
Re: TransAtlantic Cable Break
daemon@ATHENA.MIT.EDU (Deepak Jain)
Tue Jun 26 18:52:52 2007
Date: Tue, 26 Jun 2007 18:51:45 -0400
From: Deepak Jain <deepak@ai.net>
Reply-To: deepak@ai.net
To: "Chris L. Morrow" <christopher.morrow@verizonbusiness.com>
Cc: Robert Blayzor <rblayzor@inoc.net>, nanog <nanog@merit.edu>
In-Reply-To: <Pine.GSO.4.58.0706241526020.1179@marvin.argfrp.us.uu.net>
Errors-To: owner-nanog@merit.edu
> Then there's the interesting: "How do you classify 'to be dropped'
> traffic?" Simon suggests nntp or BitTorrent could be put into a lower
> class queue, I'm curious as to how you'd classify traffic which is
> port-agile such as BitTorrent though. In theory that sounds like a grand
> plan, in practice it isn't simple...
Known "high priority" traffic could just be QoSed++ and everything else
left to best-effort (or, more aptly, no effort). As Chris mentions, if
you know the endpoints, or side, or protocol, or ports, that is pretty
easy to decide. Your "everything else" bucket can be anything you
haven't specifically decided is elevated priority.
Port agile, but otherwise low-priority traffic, may or may not be
adjusted, but my guess is that *enough* of the v. large number of
strange port:strange port pairings will drop into the "everything else"
bucket for the duration of your degraded service event.
And in normal operation, with ample capacity, nothing really changes.
Deepak Jain
AiNET
