Re: IPsec & Markers

Culley, Paul wrote:

> If you were thinking of adding some kind of requirement that when doing
> IPSEC, senders MUST perfectly align and segment one FPDU to TCP segment,
> I think that would be a very difficult battle in the IETF.

Not must, but MAY (then either communicate this at start or "turn the 
knob" at boot time.

Uncle.   :-)

 > Note that the group agreed that CRC
 > checks were NOT sufficient for FPDU boundary identification.

I never found that in the archives.  I found lots of people suggesting 
this, but not a WG consensus.  I was personally convinced when Caitlin 
Bestler wrote (9 April 2004) "... there is no rule that says a valid DDP 
Segment could not be the *payload* that is being delivered within a DDP 
Segment."

That said, the archives are unreliable.  See the bottom of 
ftp://ftp.ietf.org/ietf-mail-archive/rddp/.

I also got the impression, reading the TSVWG archives, that middleboxes 
(which would be ruled out with IPsec) were the problem not the RNIC 
endpoints.  That was consistent with my personal experience from 
pre-DLSw days through "today" of sending frames in TCP and never having 
one received unaligned when both ends of the connection adhered to the 
"spec".  Hence this thread suggesting "no middlebox, no problem".

Regards,
MikeC

-- 
Michael C. Cambria

email : [email protected]
  VoIP : sip:[email protected]
   FWD : sip:[email protected]