RE: Enough on TCP vs. SCTP, please

["Douglas Otis" <dotis@sanlight.net>]

David,

I understand how you could hold that view.  To implement such a definition,
you would need to make changes to the TCP API as well as to the TCP stack to
allow messages to be marked.  There would need to be additional sorting
algorithms added to search for the first byte of a message.  It would also
be likely a one byte error would creep in with machines that do not point to
byte locations not aligned.  To solve that problem, perhaps all messages
should be frame aligned.  That could be allowed if redefining the urgent
pointer is possible. But hey, you would not need the urgent pointer then.
Sorry, but when you must re-engineer TCP to work with a SCSI specification,
you are no longer using TCP.  That is specifically why I think other means
be considered.  If you wish to see reasonable performance revise TCP?  Most
should expect without such tweaks on TCP, it will not perform well enough to
justify use with SCSI whether being clever to avoid mention or not.

Such liberties in changing TCP abound and yet Matt said he was sick of
hearing about SCTP.  Managing buffers independently is yet another example.
I personally see danger in creating unique versions of TCP to suit SCSI
whereas drive and controller manufactures may view it as their domain.
Perhaps now would be a good time to seek outside consensus on such
modifications before re-inventing TCP.  If it is true that TCP is up for
grabs, then I will be quiet as I would be wrong about this view.

Doug


> -----Original Message-----
> From: Black_David@emc.com [mailto:Black_David@emc.com]
> Sent: Friday, September 29, 2000 5:19 PM
> To: dotis@sanlight.net; Black_David@emc.com; matt_wakeley@agilent.com;
> ips@ece.cmu.edu
> Subject: RE: Enough on TCP vs. SCTP, please
>
>
> draft-wakeley-iscsi-msgbndry-00.txt says:
>
>    The proposal is pretty simple.  The first byte of every iSCSI
>    message is marked "urgent" - the TCP urgent pointer will point to
>    the first byte of the iSCSI message in the TCP segment.
>
> Douglas Otis says:
>
> > It is not reasonable to redefine the urgent pointer.
>
> Your WG co-chair says:
>
> I can't find anything in the draft that redefines the urgent
> pointer. TCP deliberately does not define exactly what the
> urgent pointer points to.  For example, I wouldn't expect
> iSCSI and telnet to use the urgent pointer in the same way.
>
> In order to continue this argument that an unacceptable change
> is being made to TCP, it is necessary to cite the text in RFC 793
> (or some other RFC that defines TCP) that would have to be changed
> if the approach in the draft referenced above were adopted.
>
> NOTE: this is *not* a statement of consensus that the approach
> in the draft should be adopted.  This is only a procedural statement
> that consideration of that approach is within scope of the WG.
>
> --David
>
> ---------------------------------------------------
> David L. Black, Senior Technologist
> EMC Corporation, 42 South St., Hopkinton, MA  01748
> +1 (508) 435-1000 x75140     FAX: +1 (508) 497-8500
> black_david@emc.com       Mobile: +1 (978) 394-7754
> ---------------------------------------------------
>