SAM-2 r6 Address Model

Ralph O. Weber ralphoweber at csi.com
Fri Jul 24 16:15:40 PDT 1998


* From the T10 (formerly SCSI) Reflector (t10 at symbios.com), posted by:
* "Ralph O. Weber" <ralphoweber at csi.com>
*
George (Ericson),

It's been awhile since this topic was on the reflector.  So, I've
cut high-level descriptions from your original comment and pasted
them in this response.  My intent is to give enough flavor of the
comment so that readers have some hope of understanding the
response.

} Issue 1:
}  Problem:
}   The SCSI model does not provide for meaningful Interconnect
}   Subsystem (i.e. bus) identification.  However, the text makes
}   various references to bus numbers or identifiers.   In the text,
}   there are various attempts to define these, but these definitions
}   are not consistent. ...
}  Solution:
}   The solution is to not attempt to label the Interconnect
}   Subsystems, but rather to provide a device relative label
}   for each Service Delivery Port of a device.
}   To Section 4.7, Object Definition 3, add 
}	Device Port = Service Delivery Port + Device Port Identifier
}	Device Port Identifier = bit<6> < [ 0|...|(2**6)-1] 
}   To Object Descriptions add: 
}	Device Port: Is a service delivery port connected ...
}	Device Port Identifier: An encoded number between 1 and ... 

For the purposes of this discussion, let's set aside the generally
held aversion for object definitions.

As I understand the proposed solution, it attempts to push the
various definitions needed for the hierarchical logical unit number
model, into the basic model for SCSI devices.  This runs counter to
the direction I see SAM going.

By missing the SCSI meetings in Portland, Maine, not only did you
miss your cookie, but you also missed the discussion wherein the
group informed me that the model description of multiple target
identifiers per target is wrong, or at least doesn't communicate
some key concepts that everybody thinks are important.

I have not worked out the details of my response to the group's
complaints, but I'm beginning to think about using the onion meta-
architecture as the basis for the SCSI architecture model; a basic
model, wrapped with a multi-port model, wrapped with a hierarchical
logical unit model.  This will mean that defining the specifics of
the hierarchical logical unit fields should be postponed to the
clause describing the hierarchical structure.  George Penokie and
several others are satisfied with the descriptions there, and I
anticipate that they will remain largely intact.

} Issue 2:
}  Problem: 
}    Bus numbering schemes for different address methods are allowed
}    to be different.  The requirement for allowing a bus (Device
}    Port Identifier) to have different values depending on address
}    method used is not spelled out in the document.  Unless needed,
}    we should settle on one model.

Having sat through this debate in a couple of SCC-2 meetings, I
am almost totally convinced that the current wording (with its
different definitions of similar fields in each of the addressing
methods) is the specific intention of the group.

Besides, the last time I proposed changing them, George reminded me
that the Logical Unit addressing method was introduced specifically
so that one faction could have an addressing method whose field
definitions differed from those of the Peripheral Device addressing
method.  That particular faction didn't like something in the
Peripheral Device addressing method (lack of access to LUNs, or 
some such) and insisted on having an address method that addressed
their requirements (if you'll pardon my pun).

} Issue 3:
}  Problem:
}    The names Peripheral device addressing method and Virtual device
}    addressing method imply the goal is to address a SCSI device
}    rather than a Logical Unit.  In fact, we always address a
}    Logical Unit through a device.

Ever since the LUNTAR bit was removed from SCSI, early in SCSI-3,
there has been no way to address a device.  One can only address
a logical unit.  So, it can be (and recently has been) argued that
describing how a logical unit is addressed is the same as describing
how a device is addressed.  I doubt that any proposed changes could
gain majority support in the working group.

} Issue 4:
}  Problem:
}    The last sentence of 4.10.4 reads: "The SCSI device located
}    within the current level shall be addressed by a BUS IDENTIFIER
}    field and a TARGET/LUN field of all zeros, also known as LUN 0
}    (see 4.10.1)."  This statement is redundant with section(sic)
}    4.10.1.  Also, it is not the SCSI Device that is addressed,
}    rather it is specifically Logical Unit zero of the SCSI Device.

N.B. All clause references and the quoted text have been updated to
match SAM-2 r8.

What you see as redundant, others see as reinforcing.  While I cannot
speak for everybody, I personally think that the concept of LUN 0 is
important enough to risk the occasional yawn from expert readers in
order to insure that everybody knows that LUN 0 must be present and
provide specific capabilities.

} Issues 5 and 6
}  Problem:
}    The 1st sentence of section 4.10.5 says a virtual device
}    "... executes command(s) using algorithms defined by a
}    configuration."   Is this a requirement or simply permission.
}    Same statement might apply to devices addressed by either
}    logical unit or peripheral device addressing method.
}  Problem:
}    In section 4.10.5, "Note 5 The virtual device might not be under
}    the control of the addressed SCSI device.  It is allowed to be
}    in an SCSI device lower in the hierarchy."

N.B. All clause references and the quoted text have been updated to
match SAM-2 r8.

These two issues concern a matter where I have a difference of
opinion with George Penokie.  I believe that SAM-2 should define
ADDRESS METHOD 01b as an address method whose function is defined
by the device model of the SCSI device to which the LUN value is
sent.  Certainly (at least to me), the SCC-2 definition of Virtual
Device addressing should take precedence over the SAM-2 definition.
Furthermore, SAM-2 in no way contains enough array controller
information to properly define ADDRESS METHOD 01b.

I take 4.10.5 to be SCC-2 words that have been watered down enough
to pass as respectable SAM-2 material, but that are otherwise
meaningless (a standardized hand-wave).  So, I'm not prepared to
climb down what I see as a senseless rat hole trying to improve the
the existing SAM-2 words.  However, I do reserve the right to take to
the working group a proposal to remove 4.10.5 from SAM-2 and identify
ADDRESS METHOD 01b as device-type specific.

Hope this helps.
Sorry to take so bloody long responding,
the interesting times have been a little too much so.

Best wishes.

Ralph...




*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at symbios.com





More information about the T10 mailing list