RE: ISCSI: Error Recovery Level 0

["Julian Satran" <Julian_Satran@il.ibm.com>]

Kevin,

Before you get too angry on this I suggest you consider all the mechanism required and what you want to achieve.
If you don't want reassignment for commands - just bring down a connection and recover the lost commands at SCSI level
you can certainly do this with what you have.

Julo

kevin_lemay@agilent.com 05/30/2002 12:26 AM Please respond to kevin_lemay

To:        John Hufferd/San Jose/IBM@IBMUS         cc:        ips@ece.cmu.edu, Julian Satran/Haifa/IBM@IBMIL         Subject:        RE: ISCSI: Error Recovery Level 0

Level 2 is not what I want at all!

In order to support level 2, I have to support level 1 plus support IO re-assignment too. I think that the data retransmission may not be worth the effort for all implementations. It is a hierarchy after all.

Level 0 is too simple - Very big hammer.
Proposed 0.5 - Simple and smaller hammer
Level 1 - Complex and for the number of expected errors, maybe not worth the effort.

You may have had a face to face, but I sure didn't. 

I don't think that the current scheme provides what I want. I understand the level 0 is for simple devices. I think that there is room for something between level 0 and 1 that is still easy to implement, but does not take the performance hit of bringing the entire session down. Modifing level 0 to only drop the troublesome connection would be OK too.

Retransmission is easy or hard depending on whether you are allowed to access the data more than once. 

Kevin Lemay

-----Original Message-----
From: John Hufferd [mailto:hufferd@us.ibm.com]
Sent: Wednesday, May 29, 2002 2:07 PM
To: LEMAY,KEVIN (A-Roseville,ex1)
Cc: 'ips@ece.cmu.edu'; Julian Satran
Subject: Re: ISCSI: Error Recovery Level 0



Kevin,
What you are asking for is Level 2, connection level recovery.  We agreed
on these levels at a Face to Face meeting, since folks ask us to make the
spec. simpler.   We all agreed on only having official levels 0,1 & 2.

Level 2 has all you want, even if you do not want to have anything to do
with level 1 (within command recovery) you can always terminate and recover
the connection.

If you attempt to make level 1 a "no-op" level, you still have to field the
SNACK on the Target Side, but you can decide to react with a connection
recovery.   Likewise on the Initiator side, you can decide to Logout the
connection instead of sending a SNACK.  Everything will work, but for the
sake of your customers you need to state that you do not do within command
recovery, even if you generally act at recovery level 2.

In any event, within command-recovery is fairly simple, and you should try
your best, then punt with a connection logout.



.
.
.
John L. Hufferd
Senior Technical Staff Member (STSM)
IBM/SSG San Jose Ca
Main Office (408) 256-0403, Tie: 276-0403,  eFax: (408) 904-4688
Home Office (408) 997-6136, Cell: (408) 499-9702
Internet address: hufferd@us.ibm.com


"LEMAY,KEVIN (A-Roseville,ex1)" <kevin_lemay@agilent.com>@ece.cmu.edu on
05/29/2002 12:14:30 PM

Sent by:    owner-ips@ece.cmu.edu


To:    "'ips@ece.cmu.edu'" <ips@ece.cmu.edu>
cc:    Julian Satran/Haifa/IBM@IBMIL
Subject:    ISCSI: Error Recovery Level 0



I have the following question on error Recovery level 0.

On page 105, v12, it says that error recovery level 0 must perform Session
recovery as described in 6.12.4.

This seems to imply that:
If I have a multi-connection session were one connection experiences a
digest error, then I must close all connections on the session, even
killing IOs over the good connection.

This is absurd! Why not just close the one connection that has a problem
and allow the IO to be restarted on one of the other live connections?
There is nothing wrong with the session, only a single connection.

I guess I am calling for a level 0.5, so that I can abort only the
troublesome connection. This would allow for a much faster IO recovery
without adding hardly any coding complexity that is required for level 1.

Comments?

Kevin Lemay