RE: iscsi : target port definition

["Jim Hafner" <hafner@almaden.ibm.com>]

Eddy, Santosh,

Marj is correct in what I'm suggesting.

If the initiator only used some of the ipaddress/tcpport (what we called
Network Portals) in a Target Portal group for one session, then it is *not*
prohibited from requesting a second session to the remaing network portals
(or any other subset of the full target portal group) within that same
target portal group *provided it used a different ISID for that session*.
The ISID RULE as stated in the draft doesn't allow it. The point being that
once the ISID is generated on the initiator, a virtual SCSI Initiator Port
is created.  That virtual SCSI Initiator Port can use any and all network
paths/bandwidth/whatever available to it to from the lower layers to
connect to a SCSI Target Port (aka target portal group).  There is no
requirement to use all the resources of the portal group.

The reason for the asymmetry has been layed out a couple of times on this
reflector.  Marj reiterates it below.   I'll rephrase it more verbosely.
The primary reason we moved from the symmetric model was to enable simpler
implementations on the target, where an implementation may have to span
multiple HW components.  The TSID RULE (which effectively is the uniqueness
of SSIDs between iSCSI Initiator and iSCSI Target) can be enforced in this
model by an implementation that is totally local to a target portal group
(partition the TSIDs across the portal groups - so they each have a piece
of the TSID namespace).


Jim Hafner


"KRUEGER,MARJORIE (HP-Roseville,ex1)" <marjorie_krueger@hp.com> on
09/14/2001 01:05:16 pm

To:   "'Santosh Rao'" <santoshr@cup.hp.com>, Jim Hafner/Almaden/IBM@IBMUS
cc:   ips@ece.cmu.edu
Subject:  RE: iscsi : target port definition



> Per the above definition, the I-T nexus target end point is that target
portal
> group, which is not the case when initiators choose to establish I-T nexi
> (sessions) to subsets of the target portal group, in order to export
multiple
> scsi paths to upper layer wedge drivers.

I think Jim is suggesting that the initiator would have to establish two
"SCSI initiator ports" to the same target portal group, which would
translate to two different ISIDs for this SCSI initiator node.  How does
that conflict with this definition?  Upper layer wedge drivers shouldn't
know the difference between two separate initiator ports providing the
multiple paths or one initiator port connecting to two different target
ports, right?

> Do you see any reasons why the definition of a target port should not be
> symmetric with the definition of the initiator port ? i.e. (iscsi target
name
> + TSID) = target port. (= both port name & port identifier). This would
more
> accurately model the target port to be the end point of the I-T nexus
(session).

Of course there are a number of combinations of possibilities for defining
what comprises the SCSI port within iSCSI.  Initially the model did specify
TSID as part of the target port identifier, but this seemed to create more
"rules" to enforce at the target side (more restrictive than necessary) and
the benefits of TSID vs target portal group were not compelling.  The idea
of target portal group as the target port endpoint seemed to provide the
necessary protection (against parallel nexus, etc) with the least amount of
enforcement rules.

Marj