RE: iSCSI: FirstBurstSize and unsolicited data

[Eddy Quicksall <eddy_quicksall@ivivity.com>]

In an earlier EMAIL this morning, I said "I think something should be said. It can be short and sweat."

 

But, I did not notice that -14 is now the final draft. So, forget what I said.

 

Eddy

-----Original Message-----
From: Julian Satran [mailto:Julian_Satran@il.ibm.com]
Sent: Tuesday, June 25, 2002 2:59 AM
To: pat_thaler@agilent.com
Cc: ips@ece.cmu.edu; nramamur@npd.hcltech.com; pat_thaler@agilent.com; rsnively@Brocade.COM
Subject: RE: iSCSI: FirstBurstSize and unsolicited data

Pat,

Do you think we should make the position of the initial data expliciit (i.e., add text) where this seems obvious?

Julo

pat_thaler@agilent.com 06/25/2002 02:30 AM Please respond to pat_thaler

To:        rsnively@brocade.com, pat_thaler@agilent.com, nramamur@npd.hcltech.com, Julian Satran/Haifa/IBM@IBMIL         cc:        ips@ece.cmu.edu         Subject:        RE: iSCSI: FirstBurstSize and unsolicited data

Robert,
 
I think there should be a restriction but it isn't as broad as requiring DataSequenceInOrder = Yes.
The necessary and sufficient requirements are (I think):
 
When immediate data is sent, it MUST begin with the first byte of the data transfer -
   rationale: there is no mechanism such as DataSN in the SCSI Command PDU to indicate position of the immediate data so it must have an implied position.
 
When unsolicited non-immediate data is sent, there is an implied R2T for the first n bytes of data
   where n = min(Expected Data Transfer Length, FirstBurstSize)
This implied R2T is to be satisfied by the immediate data plus unsolicited SCSI Data-out PDUs. As above, any immediate data must begin with the first byte of the data transfer. If DataSequenceInOrder=No, then the data in the unsolicited SCSI Data-out PDUs MAY be in any order but MUST be the bytes that satisfy the implied R2T (that is, the first n bytes of the data).
 rationale: Since the R2T is implied, it doesn't have an explicit position and must have an implicit position.
 
I don't think these requirements are articulated in the draft.
 
Pat
 
-----Original Message-----
From: Robert Snively [mailto:rsnively@brocade.com]
Sent: Monday, June 24, 2002 2:30 PM
To: 'THALER,PAT (A-Roseville,ex1)'; Nandakumar Ramamurthy; Julian Satran
Cc: ips@ece.cmu.edu
Subject: RE: iSCSI: FirstBurstSize and unsolicited data

I must have missed something in the document.  I did not see
any restriction that required EMDP = 0 (or in iSCSI-ese, DataSequenceInOrder = yes)
when immediate data was transmitted.  I would have expected such
a requirement.  If that is not the case, then any data can be transmitted
first and any data can be requested with the first (overlapping) R2T, somewhat confusing
the issue.
 
If someone can point me to that restriction, I would be delighted.
 
Bob
-----Original Message-----
From: THALER,PAT (A-Roseville,ex1) [mailto:pat_thaler@agilent.com]
Sent: Friday, June 21, 2002 9:46 AM
To: Nandakumar Ramamurthy; Julian Satran
Cc: ips@ece.cmu.edu
Subject: RE: iSCSI: FirstBurstSize and unsolicited data

Nandakumar,
 
There are no restrictions on the amount of immediate data sent other than that it must be less than MaxRecvDataSegmentLength and less than FirstBurstSize. So in the conditions you have chosen, the initiator could send any amount of ImmediateData from 4 Bytes to 8 KB. Doing so should not cause an error at the target.
 
The purpose of allowing an implicit InitialR2T is to gain efficiency by letting data start flowing without having to wait a round-trip delay for an R2T. Gaining this efficiency requires that the target know how much unsolicited data will be sent when it receives the SCSI Command PDU so that it can immediately send the first explicit R2T. The rule on sending the full FirstBurstSize (or all the data) when unsolicited data is sent in Data-out PDUs achieves this. When the target sees the SCSI Command PDU with the F bit set to 0, it knows that the first data it needs to request with an R2T starts after FirstBurstSize bytes.
 
When the SCSI Command PDU has an F bit set to 1, then the target knows that DataSegmentLength is the amound of unsolicited data being sent and it can construct its first R2T. Therefore there is no reason to restrict the amount of unsolicited data sent when only immediate data is sent (other than that it not exceed the maximums).
 
That is what is reflected in the rules.
 
Pat
 
-----Original Message-----
From: Nandakumar Ramamurthy [mailto:nramamur@npd.hcltech.com]
Sent: Friday, June 21, 2002 1:36 AM
To: Julian Satran
Cc: ips@ece.cmu.edu
Subject: Re: iSCSI: FirstBurstSize and unsolicited data

Hi,
 
I have been following the discussion on this thread.
 
I still have certain doubts regarding the
conditions I specified in my original mail.
 
The conditions are :
 
FirstBurstSize = 64KB
EDTL = 100KB ( EDTL > FirstBurstSize)
MaxRecvDataSegmentLength = 8KB
ImmediateData = Yes
InitialR2T = Yes
 
In the above case the initiator cannot send unsolicited Data-out PDUs.
Here unsolicited data(ImmediateData) < FirstBurstSize.
 
Will the target report any error in this case?
 
The modified text for Section 9.4.6.2 only refers to the cases
where unsolicited Data-outs can be sent.
 
Please clarify if I am missing something very obvious.
 
 
Thanks,
Nandakumar
Member Technical Staff
HCL Technologies, Chennai, INDIA.
http://san.hcltech.com
 
 
----- Original Message -----
From: Julian Satran
To: Eddy Quicksall
Cc: ips@ece.cmu.edu
Sent: Friday, June 21, 2002 11:41 AM
Subject: RE: iSCSI: FirstBurstSize and unsolicited data


Eddy,

I assume you meant EDTL not DSL and then the answer is yes and I again forgot to subtract  the immediate. A better formulation would be:

The target reports the "Incorrect amount of data" condition if dur-ing data output the total data length to output is greater than First-BurstSize and the initiator sent unsolicited non-immediate data but the total amount of unsolicited data is different than FirstBurst-Size. The target reports the same error when the amount of data sent as a reply to an R2T does not match the amount requested.

Julo

"Eddy Quicksall" <Eddy@Quicksall.com> 06/21/2002 12:22 AM Please respond to "Eddy Quicksall"

To:        Julian Satran/Haifa/IBM@IBMIL        cc:                Subject:        RE: iSCSI: FirstBurstSize and unsolicited data

Isn't this saying that if DSL > FirstBurstSize, it would be incorrect to send non-immediate unsolicited which is not equal to FirstBurstSize?
 
The target reports the "Incorrect amount of data" condition if dur-ing data output the total data length to output is greater than First-BurstSize, but the initiator sent an amount different than FirstBurstSize of unsolicited non-immediate data or the amount of data sent as a reply to an R2T does not match the amount requested.
 
Eddy