RE: iSCSI: No Framing

["ERICKSON,SHAWN (HP-Roseville,ex1)" <shawn_erickson@hp.com>]

Not trying to be obtuse but how does it help if only the NIC vendors
implement it? (I know you didn't precisely say that)

It sounds like we may end up with NIC vendors implementing it and storage
vendors ignoring it or some mix. In the end we will have a fracture in the
market place and/or all vendors will be forced to implement FIM to satisfy
customer demands ("my NIC supports FIMs, you don't, things are SLOW, you
must be at fault", etc.).

To me it sounds like FIMs needs to be SHOULD/MUST or not in the standard at
all (moved to a secondary document).

I personally would like to see FIM (and COWS) moved from the standard to a
secondary document but that is just my gut level opinion on this issue. I do
think FIM blurs the layering of TCP/IP, which is not normally desirable.

Regardless it would be worthwhile for a document describing how to fire
iSCSI over SCTP or other IP backed transports that are better suited for
iSCSI style data and high speed/long distance links. (Please excuse my
ignorance is such drafts already exist)

-Shawn

-------------------------------------------------------
 Shawn Carl Erickson

> -----Original Message-----
> From: Jim Pinkerton [mailto:jpink@microsoft.com]
> Sent: Tuesday, February 05, 2002 6:53 AM
> To: John Hufferd; somesh_gupta@silverbacksystems.com
> Cc: Mukund, Shridhar; ips@ece.cmu.edu
> Subject: RE: iSCSI: No Framing
> 
> 
> 
> In reviewing the mail alias, and after having many conversations with
> various folks, I would like to retract my prior opinion on ripping out
> markers. It appears to me as though _every_ NIC vendor is in 
> support of
> markers. 
> 
> To me the evidence in support of keeping in "Sync and Steering with
> Fixed Interval Markers" is pretty compelling (now that I've been
> re-educated). Without going into proprietary issues - how often do NIC
> vendors agree unanimously on something? Also, a software 
> implementation
> of markers is tractable.
> 
> I have not heard a single NIC vendor in support of COWS, and I
> personally don't support COWS either. Thus I would recommend that COWS
> be "ripped out".
> 
> 
> Jim