RE: NOP-Out clarification

[julian_satran@il.ibm.com]

Ashish,

TCP already guarantees data delivery. Data Numbering was there to enable
long running input operation to recover faster (repeating only unacked data
after a restart).  And many members of this list (not all) felt it was
adding to much complexity for what it bought.

As for scoreboarding in all SCSI protocols the target is supposed to do its
job and the delivery subsystem to deliver. The initiator is not supposed to
do scoreboarding (it can do for completeness but it is not required to) and
has to report exceptions (including counters for overrun/underrun) as they
where conveyed by the target.

Julo

"Choubal, Ashish" <ashish.choubal@intel.com> on 19/01/2001 02:07:46

Please respond to "Choubal, Ashish" <ashish.choubal@intel.com>

To:   Julian Satran/Haifa/IBM@IBMIL
cc:
Subject:  RE: NOP-Out clarification




Hi Julian,

Scraping Data Numbering will affect how single-initiator-command - multiple
data response PDU
case is handled. My understanding was that the data numbering mechanism
allowed the
initiator not to do any score-boarding and put the burden on the target to
guarantee that
all data requested by initiator was delivered - i.e. target numbers the
each
of response
data PDU and initiator acknowledges it received the PDU via a NOP. When
target receives
acknowledgements for all the DataRN's it sent, then it sends a status info
to complete the
command.

What am I missing?

-Ashish Choubal


 -----Original Message-----
From:     julian_satran@il.ibm.com [mailto:julian_satran@il.ibm.com]
Sent:     Thursday, January 18, 2001 3:51 PM
To:  ips@ece.cmu.edu
Subject:  Re: NOP-Out clarification



JP,

The consesus at the intermediate meeting in Orlando was to drop Data
Numbering.
If you have strong object here is your last chance to object.

Also Data Numbering was meant only for incoming data and it is per command
("less than a task" if you take in account linked commands).

Regards,
Julo


Please respond to Raghavendra Rao <jp.raghavendra@sun.com>

To:   ips@ece.cmu.edu
cc:
Subject:  NOP-Out clarification





My reading of NOP-out as used for data-in acknowledgements is that
it is for the entire task - Not per R2T since I don't see offset/length
in the NOP-out header. Is my assumption correct ? It would be nice
if the draft can be more explicit.

Thanks.

-JP