Re: silent dropping of TCP segment

On Mar 25, 2004, at 7:13 PM, Culley, Paul wrote:

> The use of "committed" does not imply that a segment is acknowledged,
> only that this is the one that will be kept (a future segment with the
> same sequence numbers will not be passed to MPA).
>
> Any implementation that intends to drop segments based on CRC errors,
> must be sufficiently integrated with TCP to insure that such segments
> are not acknowledged; and thus will be re-transmitted.
>
> I don't think that just deleting the MUST is sufficient, since there 
> are
> other implications, (and I'm not sure I remember them all).  I'm open 
> to
> suggestions...
>
> Paul R. Culley
> HP Fellow
> 281-514-5543

The way I've interpreted those clauses is that the TCP Layer
should only present a segment for placement that it has
*decided* to ack, and (as you state) will not be accepting
a further retransmission of.

The MPA driven "drop" essentially *reverses* that decision.

You could quibble about whether a decision that can be
overridden by MPA qualifies as "committed" or not. But
I believe the concept is fairly simply, it's just a matter of
agreeing on the right terms, or agreeing that the words
that are already there are compatible with what was meant.