Silently dropping QoS marked packets on the greater Internet

Mark Tinka mtinka at globaltransit.net
Fri Sep 9 05:16:18 UTC 2011


On Saturday, September 03, 2011 12:02:03 AM 
Valdis.Kletnieks at vt.edu wrote:

> Except you can't actually *guarantee* that QoS works
> every packet, every time, during congestion even within
> the same network. Remember - QoS is just a marking to
> shoot the other guy first.  If a link ends up
> overcommitted with QoS traffic, you're still screwed. 
> And there's a second-order effect as well - if your net
> is running sufficiently close to the capacity edge that
> QoS actually matters, there's probably other engineering
> deficiencies that are just waiting to screw you up.

Agree.

What we've seen (and I suppose what the design philosophy 
suggests) is that so-called Priority traffic has the highest 
chance of survival during times of evil. But then again, 
depending on just how saturated the port queues are, even 
Priority traffic can get dropped due to lack of buffers - 
that is if it hasn't already been caught by policers that 
tend to go along with Priority queues.

> Is the story I've heard about people managing to saturate
> a link with QoS'ed traffic, and then having the link
> drop because network management traffic was basically
> DoS'ed, apocryphal, or have people shot themselves in
> the foot that way?

This sounds like a hacked attempt to get management to 
approve that 40Gbps upgrade :-).

Mark.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
URL: <http://mailman.nanog.org/pipermail/nanog/attachments/20110909/aad0143e/attachment.sig>


More information about the NANOG mailing list