RE: iSCSI: DLB's [T.6] 2.3 iSCSI Session Types

[Black_David@emc.com]

Julian,

 

Your "at its own risk" is an invitation to interoperability problems -

if Discovery sessions are meant to be fixed functionality, then an

Initiator that attempts to exceed that functionality needs to get

told NO - otherwise we're going to see READ commands on Discovery

sessions to get additional configuration information, and then

task management, and ... this neat stuff will only work right

when the same implementation is on both ends of the connection.

 

I have a separate reply to Ayman's message coming that contains

an attempt at requirements text, but it's longer than I would like.

 

Thanks,

--David

 

-----Original Message-----
From: Julian Satran [mailto:Julian_Satran@il.ibm.com]
Sent: Tuesday, July 09, 2002 1:40 AM
To: Black_David@emc.com
Cc: aghanem@cisco.com; ips@ece.cmu.edu; owner-ips@ece.cmu.edu
Subject: RE: iSCSI: DLB's [T.6] 2.3 iSCSI Session Types

David,

The original MAY was perfectly appropriate and so is the MUST.
If you say that you have targets that MAY not support anything else than .. the an initiator attempting to use anything else
would do it at its own risk.
Your preference for MUST puts an additional mandatory burden to targets to CHECK (all targets) that nothing else
is sent.

It is strict where none of us thought that we have to be.

From an initiator POW they are equivalent.
For a target it requires more work.

I would favor the way we where ... but that may be all age related.

Julo

Black_David@emc.com Sent by: owner-ips@ece.cmu.edu 07/08/2002 05:18 PM Please respond to Black_David

To:        aghanem@cisco.com, ips@ece.cmu.edu         cc:                 Subject:        RE: iSCSI: DLB's [T.6] 2.3  iSCSI Session Types

Ayman,

Something needs to be cleaned up here, as the current text appears
to allow all types of iSCSI PDUs on a discovery session.  I didn't
intend to restrict a discovery session to one Send Targets followed
by a logout (i.e., it could be kept open with the initiator periodically
sending a new Send Targets to see if anything has changed), but I
did intend to forbid SCSI commands, task management, etc. on Discovery
sessions.  Is that reasonable, or are there additional types of iSCSI
PDUs that you want to see allowed for new device notifications?

Thanks,
--David

> -----Original Message-----
> From: Ayman Ghanem [mailto:aghanem@cisco.com]
> Sent: Sunday, July 07, 2002 12:41 PM
> To: ips@ece.cmu.edu
> Subject: RE: iSCSI: DLB's [T.6] 2.3 iSCSI Session Types
>
>
> I prefer leaving this as "MAY" for implementations that want
> to support new
> device notifications. There was a discussion on whether
> discovery sessions
> should be long-lived or not. Using MAY allows both without
> breaking any
> thing.
>
> -Ayman
>
> > [T.6] 2.3  iSCSI Session Types
> >
> >       b)  Discovery-session - a session opened only for
> target discov-
> >       ery; the target MAY accept only text requests with
> the SendTar-
> >       gets key and a logout request with reason "close the session".
> >
> > Change "MAY" to "MUST", and say that other requests MUST be
> rejected.
> >
>