> Date: Mon, 22 Mar 1999 16:57:20 -0500
> To: Tim Salo <[EMAIL PROTECTED]>, [EMAIL PROTECTED]
> From: Steve Meuse <[EMAIL PROTECTED]>
> Subject: Re: VC vs. DG modes
>
> >ECN tries to provide TCP better feedback about congestion than merely
> >dropping packets, (or source quench). (You should probably look at the
> >random early detection (RED) stuff as well.)
>
> I've allways thought that RED and AX.25 would work well together. However,
> I think the "I would have dropped you, slow down" communication at 1200
> baud would cause just as much congestion as the congestion that you are
> trying to alleviate.
> [...]
ECN uses [the last] two previously reserved bits in the IP header. (So,
why have we been sending bits that are never set over 1200 bps links?)
A router can use RED policies for congestion detection to set the
"congestion experienced" bit in the IP header. An objective of RED
is to avoid congestion by getting senders to slow down before congestion
(buffer overflow) occurs. ECN provides RED-capable routers another
method of informing the hosts of impending congestion, (i.e., set the
"congestion experienced" bit rather than dropping the packet).
Does anyone have any empirical experience with TCP continuing to probe
a path after the path is full, or are most amateur packet paths so bad
that TCP never fills the pipe?
-tjs
