[3371] in linux-net channel archive
Re: Traffic shaper
daemon@ATHENA.MIT.EDU (mike@lserv.conexio.co.za)
Thu Jun 20 20:42:26 1996
From: <mike@lserv.conexio.co.za>
To: Jon Lewis <jlewis@inorganic5.fdt.net>
cc: Mike Kilburn <mike@lserv.conexio.co.za>, Alan Cox <alan@cymru.net>,
linux-net@vger.rutgers.edu
In-reply-to: Your message of "Thu, 20 Jun 1996 13:50:30 -0400."
<Pine.LNX.3.91.960620134625.15516Z-100000@inorganic5.chem.ufl.edu>
Date: Thu, 20 Jun 1996 14:05:49 -0500
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Thu, 20 Jun 1996 19:49:14 +0200
From: Mike Kilburn <mike@conexio.co.za>
> I'm pretty sure you will have to drop some frames...or die trying not
> to. We had this problem a few months back where Iphone or CuSeeMe
> users sessions would kill the terminal server because the other end of
> their link was spitting out UDP packets much faster than the modem could
> accept them, and eventually the system ran out of memory and crashed.
>
Yes I know. But I did not want to be to quick to drop in the hope that
the effictively delayed ack from the receiver would cause a 'flow control'
effect at the source thus avoiding retranmission. It has now
been pointed out to me that this will have a different effect on protocols
like TCP. Something evil like Iphone would of course not stop so it would
be a wasted effort to buffer that too I guess.
