Re: Towards Consensus on TCP connections

[Rod Van Meter <rdv@cips.nokia.COM>]

Sigh, I've been way too busy to contribute significantly here, and there's 
nowhere I'd rather be :-(.  We had MANY of these same arguments at Quantum in 
1998 and 1999; they're endlessly fascinating but can really only be resolved 
by implementation and comparison.

I've known about GPP for years, but my memory fades.  I was told by a member 
of T10 at one point that it was really put together and driven by someone in 
particular, and when his application failed to take off, was more or less 
abandoned.  I don't recall who, or what the application was.

SCSI over ST and ST over IP offer a clear path today, but I don't think it's 
adequate.  I'm glad to see the full iSCSI work going on.

I'm in favor of keeping FC over IP out of the equation; I'm opposed to it, 
anyway, but I think this group should continue to concentrate on iSCSI as 
currently understood.

It's a little late (I think we're past the "TCP is too expensive" argument for 
the moment), but I happened across this URL this morning in the references of 
Wright & Stevens, v.2.

http://www.kohala.com/start/vanj.93sep07.txt

TCP processing in 30 instructions, according to Van Jacobson.

I hand-counted the Linux 2.2 stack, and came up with about 200 instructions, 
not counting checksum, but including the interrupt handler, ethernet 
processing, IP, and TCP, I believe.  I'll see if I can dig up my notes and 
post them here.

I'm in favor, btw, of multiple TCP streams.

Anyway, I'm hoping to contribute some to this discussion from this point 
forward.  Sorry I haven't been involved much to date.

		--Rod