Re: Urgent as Framing Hint?

["Randall R. Stewart" <randall@stewart.chicago.il.us>]

"David P. Reed" wrote:
> 
> At 06:19 PM 12/1/00 +0100, Luigi Rizzo wrote:
> 
> >not that i don't like the idea, but this way any "standard" receiver
> >will not understand the new checksum and drop the packet. Net effect
> >is that you can only talk to systems that understand the new checksum
> >style unless you have some way to negotiate thhis feature.
> 
> One way to negotiate this feature is to use it and see what happens.  The
> basic approach would be to send a datagram containing the variant checksum
> as the first datagram in the TCP stream.  If the receiver discards the
> frame, it will never acknowledge it, even upon retries.  If it receives an
> acknowledgement, also variant-marked, the sender and receiver conclude the
> negotiation.  If not, the sender can then drop the connection and indicate
> to the TCP client that the variant-marking of frames is not supported, and
> a default could be used.
> 
> >If you
> >don't, i fail to see how this is different from having a totally
> >different protocol which happens to have the same ID as TCP (perhaps
> >to get through firewalls ?).
> 
> Well, TCP-Reno is  by this definition a "different protocol which happens
> to have the same ID as TCP", right?
> 
> Please be aware, I make this suggestion because the idea seems to be
> helpful and low cost, given the constraints that the iSCSI folks seem to be
> laboring under.  I leave advocacy and discussion of the impacts to others,
> esp. those whose needs benefit and those who know of adverse impacts.  But
> I hope it might be a way out of a heated conflict where all sides have some
> legitimate concerns.
> 

David:

I do like your idea, it provides a "quick" simple solution. The real
issue then goes back to a earlier problem mentioned in this debate.
How long would it take for the standard "O/S" vendor TCP's of the
world to adopt this and get it into production releases? I would
imagine quite some time... 

I think this is true of any solution that can't use TCP "out of the
box",
has it exists in the internet today... If the iSCSI framing folks are
willing to wait for 1-2 years for massive deployment (assuming they
can get all the standard "O/S" vendors to agree to roll it out) then
it is a very valid and useful proposal :) I have no idea what the
requirement for time frame is so I can't judge this...

R




> - David
> --------------------------------------------
> WWW Page: http://www.reed.com/dpr.html

-- 
Randall R. Stewart
randall@stewart.chicago.il.us or rrs@cisco.com
815-342-5222 (cell) 815-477-2127 (work)