Re: revision RFC 2001 (draft-ietf-tcpimpl-cong-control-00.txt)

[Vern Paxson <[e-mail]>]

> In the draft, the sentence "Specifically, an ACK MUST be generated for
> every second full-sized segment.", is not sufficient to assure correct
> operation because the receiver can not identify segments which the
> sender considers to be full sized.

Good point.

> I suggest changing the paragraph to note that the classical delayed ACK
> specification predates pMTU discovery, and is no longer sufficient.
> Use the simpler, more explicit:
> "An ACK MUST be generated for every second segment, regardless of
> size.  (replacing "second" by.....?)
> 
> Yes, all(?) existing implementation are non-conformant, and yes all(?)
> can be made to misbehave on real paths.

This is too big a change, IMHO.  While acking every second segment
regardless of size appears sound (see my next message), it's not existing
practice.  Van likes the notion of ack-every-other (he likes even more
ack-every :-), but agrees that there's insufficient experience with it
to make it required (it's already allowed).  He points out that it could,
for example, stress SWS-avoidance algorithms that today don't get stressed
much, and these could break.

So I think the way to deal with this is keep the current wording but
add a note stating that the requirement is in terms of what the receiver
can deduce about the sender's full-sized segments, and that the receiver
should allow for possible changes in PMTU when reckoning this.

		Vern