Re: [Tsvwg] Fwd: I-D ACTION:draft-ludwig-tsvwg-tcp-eifel-alg-00.txt

At 16:36 23.11.00, Lloyd Wood wrote:
>On Thu, 23 Nov 2000, Reiner Ludwig wrote:
>
> > I would very much appreciate to get some feedback on this mailing list
> > concening draft-ludwig-tsvwg-tcp-eifel-alg-00.txt.
>
>wouldn't D-SACK information make Eifel flag negotiation redundant,
>while also allowing simplifying timestamp handling somewhat?

Unfortunately, D-SACK does not help you to reliably detect a spurious 
timeout as soon as it is required to avoid the go-back-N. Consider,
a) a timeout in response to a genuine packet loss, and
b) a spurious timeout.

The first ACK that arrives back at the sender acking the retransmit will 
"look the same" in both cases; with or without D-SACK. Thus, the sender 
would have no way to tell whether that ACK was triggered by the original 
transmit or the retransmit, i.e., the sender would not know (at that point) 
whether the last timeout was spurious or not.

>I think the most valuable part of the flag approach is:
>
>    The receiver SHOULD send an immediate ACK with the Xmit-Echo flag set
>    in response to an incoming data segment that has the Xmit flag set.
>
>since then the sender has the chance to demand real path delay without
>the arbitrary delayed acknowledgement timer getting in the way at the
>receiver, and to compensate for long path delays for e.g. telnet by
>setting the flag a lot.

You should also read the "security consideration" section. That forbids 
such abuse of the Xmit flag. The Xmit flag MUST only be set for 
retransmits. It MUST NOT be set in any other case.

///Reiner

_______________________________________________
tsvwg mailing list
tsvwg@ietf.org
http://www1.ietf.org/mailman/listinfo/tsvwg