Ming,

Reservations (not persistent reservations) were defined long before that was any such thing as multiple port devices and as such is not well suited for use is such environments. That is one of the reasons reservations no longer defined in the current SCSI specifications and were replaced with persistent reservations.

In the process of defining persistent reservations is was not clear in the standards as to what the reservations were associated with. At the same time the SCSI architecture was moving from a parallel based architecture to a serial based architecture. That change required a clear delineation between ports and devices that was not previously necessary. The end result is that the I_T nexus became a port to port definition and that made all reservations port to port.

I would suggest the only reasonable solution is the use of persistent reservations in environments that contain multi-ported SCSI devices.
 
Bye for now,
George Penokie

Dept 9A8 030-3 A410
E-Mail:    gop@us.ibm.com
Internal:  553-5208
External: 507-253-5208


Ming Zhang <blackmagic02881@gmail.com>

01/02/2007 08:38 AM
Please respond to
blackmagic02881@gmail.com
To
George Penokie/Rochester/IBM@IBMUS
cc
Arne Redlich <agr@powerkom-dd.de>, Juhani Rautiainen <jrauti@iki.fi>, owner-t10@t10.org, "Ross S. W. Walker" <rwalker@medallion.com>, Steffen Plotner <swplotner@amherst.edu>, t10@t10.org
Subject
Re: question about reservation and persistent reservation





Hi George

Thanks for answering my question.

>From iSCSI RFC (http://www.faqs.org/rfcs/rfc3720.html)
I_T nexus: According to [SAM2], the I_T nexus is a relationship
    between a SCSI Initiator Port and a SCSI Target Port.  For iSCSI,
    this relationship is a session, defined as a relationship between
    an iSCSI Initiator's end of the session (SCSI Initiator Port) and
    the iSCSI Target's Portal Group.  The I_T nexus can be identified
    by the conjunction of the SCSI port names; that is, the I_T nexus
    identifier is the tuple (iSCSI Initiator Name + ',i,'+ ISID, iSCSI
    Target Name + ',t,'+ Portal Group Tag).

Why? In iSCSI, if one iSCSI initiator setup 2 or more sessions with
_same_ target, i could not see any negative effect that commands can not
send through all sessions to same target from same initiator.

Or one step ahead, when SCSI spec is defined, why reservation need to be
nexus based (between initiator port and target port), not between
initiator and target, for multiple port situation?

Ming


On Tue, 2007-01-02 at 08:27 -0600, George Penokie wrote:
>
> Ming,
>
> Reservations and persistent reservations are both I_T nexus based. The
> question that has to be answered is; What has the iSCSI protocol
> defined as the initiator port (i.e., the I) and the target port (i.e.,
> the T). From SAMs point of view as long as the connection is between
> the same I and T that holds the reservation then the command is valid
> regardless of the actual physical path that is used (i.e., there can
> be any number of physical paths between a single initiator port and a
> single target port).
>
> Bye for now,
> George Penokie
>
> Dept 9A8 030-3 A410
> E-Mail:    gop@us.ibm.com
> Internal:  553-5208
> External: 507-253-5208
>
>
> Ming Zhang
> <blackmagic02881@gmail.com>
> Sent by: owner-t10@t10.org
>
> 12/28/2006 10:21 AM
>          Please respond to
>      blackmagic02881@gmail.com
>
>
>
>
>                To
> t10@t10.org
>                cc
> Arne Redlich
> <agr@powerkom-dd.de>, Juhani Rautiainen <jrauti@iki.fi>, Steffen Plotner <swplotner@amherst.edu>, "Ross S. W. Walker" <rwalker@medallion.com>
>           Subject
> question about
> reservation and
> persistent
> reservation
>
>
>
>
>
>
>
>
> * From the T10 Reflector (t10@t10.org), posted by:
> * Ming Zhang <blackmagic02881@gmail.com>
> *
> Hi All
>
> We try to add reservation support to an iSCSI target and have some
> questions about reservation.
>
> Reservation is per nexus or per initiator? persistent reservation is
> per
> nexus or per initiator?
>
> Our dilemma is like this. between an iSCSI initiator and target,
> multipath IO can (1) has multiple sessions over to same target via
> different physical NICs and paths. or (2) has one session with
> multiple
> connections, with each connection over different paths. Base on iSCSI
> RFC, (1) will have multiple I_L nexus while (2) only have 1 nexus.
>
> Then for reservation support via RESERVE_6/10, if one initiator
> reserve
> one LU via one session, can this initiator send WRITE or other
> commands
> via another session?
>
>
> Thanks
>
> Ming
>
>
> *
> * For T10 Reflector information, send a message with
> * 'info t10' (no quotes) in the message body to majordomo@t10.org
>
--
http://blackmagic02881.wordpress.com/