RE: iSCSI: Markers

["Glenn Dasmalchi" <glennd@chelsio.com>]

I agree with Steph's conclusion on this one. The PDU Containment
Property of TUF is a good fit for high-bandwidth direct placement
receive hardware. If COWS *with* TUF-style segmentation is deemed
impractical, then I think it's better to have PDU containment in TCP
segments (with random key/length headers) than fixed interval markers.

I understood David's earlier posting about not allowing normative
references to experimental drafts (i.e. TUF), but I don't understand how
this precludes using the technique as the iSCSI framing mechanism. If we
decided it was the right technique for iSCSI, couldn't we just lift the
content and put it in the iSCSI draft?

-Glenn

> -----Original Message-----
> From: Stephen Bailey [mailto:steph@cs.uchicago.edu]
> Sent: Wednesday, January 09, 2002 5:34 AM
> To: ips@ece.cmu.edu
> Subject: Re: iSCSI: Markers 
> 
> 
> > As for COWS, it's clearly nasty.  Even if you're doing CRC, 
> it adds a
> > significant increment to the processing cost.
> 
> Unfortunately, all the choices are nasty.  Long time IETFers would
> probably make the point that this is an indication we are trying to do
> the wrong thing and should just use a message-based transport :^)
> 
> That said, one position that many people thought was the least nasty
> was key/length-based TUF (with TCP sender segmentation support), or
> nothing at all (== full reassembly).  A smaller number of `key'
> participants felt that an intermediary framing solution for stock TCP
> senders was necessary.
> 
> We proposed COWS-based TUF to try to bring harmony to the spheres, but
> I'm not hearing the pleasant resonance (piano tuning is an emperical
> process, right?), so perhaps we didn't get it quite right.
> 
> Absent any unification of sender segmentation-based and stock TCP
> framing (perhaps 0-touch sends ARE too much to give up), I'm strongly
> in the sender segmentation-based framing (TUF) or nothing at all camp.
> 
> Steph
>