[ previous ] [ next ] [ threads ]
 From:  "Michael Pope" <spotnruby at gmail dot com>
 To:  m0n0wall at lists dot m0n0 dot ch
 Subject:  Re: [m0n0wall] Traffic Shaping on Network Perimeter Device
 Date:  Wed, 13 Aug 2008 11:17:11 -0400
Thank you all for your explanations. I think I now have a better grasp of
the reasoning behind traffic shaping.


On Tue, Aug 12, 2008 at 6:48 PM, Brian Lloyd <brian dash wb6rqn at lloyd dot com> wrote:

> That's why mostly, shaping is done on the perimeter, where the LAN traffic
>> exceeds the capabilities of the WAN line.  Any shaping done should
>> feedback
>> to the internal nodes to slow down (I don't recall how traffic shaping's
>> implemented), and in essence, help control the flow through the pipe.
> ICMP source quench was supposed to solve the problem by letting an
> intermediate system send a message to an end system and get it to throttle
> its traffic. Turned out that it ended up making the traffic level oscillate.
> (While not particularly useful, it was fun to watch.)
> Today the solution is to just throw packets away and let TCP cope. Most TCP
> implementations assume that lost packets are due to network congestion and
> close the transmit window. It is surprising just how well TCP will converge
> on proper throughput.
> (I used to design routers in a previous life.)
> Brian Lloyd
> Granite Bay Montessori School          9330 Sierra College Bl
> brian AT gbmontessori DOT com          Roseville, CA 95661
> +1.916.367.2131 (voice)                +1.791.912.8170 (fax)
> PGP key ID:          12095C52A32A1B6C
> PGP key fingerprint: 3B1D BA11 4913 3254 B6E0  CC09 1209 5C52 A32A 1B6C