Re: RF flag

art@sage.acc.com Tue, 05 December 1989 23:23 UTC

Received: from decwrl.dec.com by acetes.pa.dec.com (5.54.5/4.7.34) id AA23536; Tue, 5 Dec 89 15:23:30 PST
Received: by decwrl.dec.com; id AA16012; Tue, 5 Dec 89 15:23:23 -0800
Message-Id: <8912052323.AA16012@decwrl.dec.com>
Date: 5 Dec 89 15:22:00 PST
From: art@sage.acc.com
Subject: Re: RF flag
To: "mtudwg" <mtudwg>

>It is understood that Congestion is a Really Big Problem, but it is
>far from agreed that the DEC bit is the solution.  Jacobson's approach
>of discarding packets as a congestion "signal" has the advantages of
>(a) immediately reducing the number of packets in the network, and
>(b) working with existing host implementations.

Any time that you push the net to the point of dropping packets, the
total useful throughput drops.

I think that Van might even agree that the DEC Bit has some advantages.
The DEC Bit tends to detect congestion when it first appears and attempts
to signal this to the end systems before the congestion gets out of hand.
Van's algorithm uses the only currently available signal, congestive
loss.  In this mode, the routers are forced to operate near what DEC
calls the "cliff".  Any surge in load at this operating point causes
large losses.  The DEC Bit attempts to keep the router at an operating
point (the "knee") which can still cope with traffic fluctuations.
This point also optimizes the throughput/delay ratio.  Also the DEC Bit
is better at discriminating between congestion and real loss.

>Respectfully yours,
>Steve

+-----------------------------------------------------------------------+
|	Art Berggreen		Advanced Computer Communications	|
|	<art@sage.acc.com>	Santa Barbara Street			|
|	(805)963-9431		Santa Barbara, CA 93101			|
+-----------------------------------------------------------------------+