RE: iSCSI : default iSCSI mode page settings - Consensus call

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

Robert,

I agree. Per LU parameters have little or no relation to iSCSI.

Regards,
Julo

Robert Snively <rsnively@Brocade.COM> Sent by: owner-ips@ece.cmu.edu 02-10-01 17:36 Please respond to Robert Snively

To:        ips@ece.cmu.edu         cc:                 Subject:        RE: iSCSI : default iSCSI mode page settings - Consensus call

Folks,

In an offline discussion, we have picked at a thread
central to this fabric.  That is basically the view that
one has of the structure of targets.

------------------   Note sent out a day ago  ---------------

Traditional network environments talk from one big host
memory to another big host memory.  In the storage space, the
memory in the target is actually distributed between whatever
caching and preprocessing space the target platform may have
and the space held by each logical unit.  In some cases (RAIDs)
the distribution is heavily in favor of the target platform
cache.  In others (tapes), individual logical units may have
100s of Megabytes of buffering which are not available to other
logical units.  If you have mixed types of logical units or
if you have the memory distributed unevenly among the logical
units, different values are normal among different LUs.

Different sessions to the same LU may also have different
properties.  In particular, the LU may give away resources
freely for the first few sessions, but more sparingly for
subsequent sessions.  Alternatively, the LU may share the
resources dynamically, providing UNIT ATTENTION indications
if it must change the values.

The iSCSI target is merely a handy portal to the logical
units and device servers, which really run the show.  While
many targets are simplified, sharing parameters across LUs
and sessions, there is no requirement for that.  Sophisticated
devices may have individual parameters for each LU and
session.  iSCSI is trying hard to ignore the differences
between storage and networking devices, but the differences
are justified and genuine.

----------------   Some of the questions that triggered the note  -----

> Are you saying that these parameters need to be able to have different
> values for each LU among several LUs of single target in a single
> session?
> 
> Also would different sessions to the same LU all need to use the same
> value?

---------------  And a succinct summary of the thoughts  ---------------

If I understand correctly, in your view, such parameter negotiations
should be per initiator and per LU.  Parameters which are unique per
logical unit can be set (or "negotiated") *only* via mode pages.
Whether or not changing the value for one logical unit changes it for
others would be target end implementer's choice, subject to appropriate
parameter change notifications to the initiator(s).  The values
negotiated during login would be per initiator and per target
default/initial values to be used when per LU mode pages have not been
subsequently modified by this initiator.

I am not reading that you think these values should not be settable at
the per session level.  Only that you believe there is a real need to be
able to change them at the per LU level so that the per session values
can be over-ridden for specific LUs.  Changing the value for one LU via
the mode pages would not have to change it for other LUs in this session
nor for other sessions to this LU.

Is that about right?

---------  My answer would be yes  -------------------------------------