Can one LU be in multiple devices?

George Ericson GEricson at TownISP.com
Tue Oct 5 03:45:24 PDT 2004


* From the T10 Reflector (t10 at t10.org), posted by:
* "George Ericson" <GEricson at TownISP.com>
*
This is a multi-part message in MIME format.

------=_NextPart_000_002E_01C4AAA6.E53C8BE0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: 7bit

Jim, George,

 

When I invented the LU mapping concept back in 96/97, the same domain
issue was present.  I interpreted the domain as enforcing naming rules
on LUNs and port IDs (known then as target IDs).   Note that even
without virtualization, a multi-ported target device can participate in
more than one SCSI domain.  I did not interpret the domain as enforcing
any rules on the LU itself.  The key to LU Mapping was to virtualize the
domain, initiator ports and target ports (and thus initiator and target
devices).  Particular (virtual) ports are tied to a particular (virtual)
domain and collectively must conform to SAM-x.  

 

Hope this helps

George

 

  _____  

From: owner-t10 at t10.org [mailto:owner-t10 at t10.org] On Behalf Of Jim
Hafner
Sent: Monday, October 04, 2004 4:23 PM
To: George Penokie
Cc: Mallikarjun C.; Reuter, Jim; Julian Satran; KRUEGER,MARJORIE
(HP-Roseville,ex1); Paul.Vonbehren at sun.com; T10 Reflector
Subject: Re: Can one LU be in multiple devices?

 


George, 

In the first place, I'd never try to convince you that you are "wrong"
:-).   

And I believe that what you say is the correct interpretation of the
current SAM-3.   

And I have no intention of trying to change this even for SAM-4 (I have
no vested interest in this -- at least not at this time).  I'm really
trying to help out Mallikarjun and friends who brought the question to
me (and I apparently answered wrong!). 

But, I'm not sure you answered my last questions.  The key to me is this
one: 
** Why does the domain impose this LU constraint on the model? ** 
In other words, what inherent property of the domain requires the
constraint for model consistency? (Or is this just an historical fact?).

Or from a different viewpoint, what feature does this constraint provide
within a domain?  And how is this feature handled across domains? (Or is
it simply that the standard only addresses "within a domain" and doesn't
address "cross-domain" issues?) 

Thanks, 

Jim Hafner/Almaden/IBM
Tel: (408) 927-1892, Fax: (408) 927-3030 (t/l 457)
Email:hafner at almaden.ibm.com 




 

George Penokie 

09/30/2004 01:15 PM 


        To:        Jim Hafner/Almaden/IBM at IBMUS 
        cc:        "Mallikarjun C." <cbm at rose.hp.com>, "Reuter, Jim"
<james.reuter at hp.com>, Julian Satran <Julian_Satran at il.ibm.com>,
"KRUEGER,MARJORIE (HP-Roseville,ex1)" <marjorie.krueger at hp.com>,
Paul.Vonbehren at sun.com, T10 Reflector <t10 at t10.org> 
        Subject:        Re: Can one LU be in multiple devices? Link
<Notes://882563A90059E2B4/5E502A1BAAAF40CA85256197006C1A32/8E44E702F0884
F4B85256F1E007533EB> 



Jim, 

The constraint is that one logical unit cannot span multiple target
devices when both those targets devices and the logical unit are in the
same domain. (This is a very simplified statement that is described in
much more gory detail in my previous note. 

I believe your example and the statement after is correct. 

You ask why it is this way. My question is what's wrong with the current
definition. As far as I know, that are now cases where it causes a
problem. Also, allowing it would cause problems in identifying and
locating specific target devices. 

Before you start trying to convince my I am wrong, STOP, it will do no
good. The standard is the way I have described it and SAM-3 is done. To
change SAM-4 you have to convince the CAP working group, not me,  not
only that a change is needed but you will have to write up all the
changes to the standards that would be required to implement the change.


Consider these notes an education as to how it is defined today and go
|from there, if you dare. 

Bye the way the t10 reflector has been dead for most of this thread.
When it comes up I will try to make sure this is all posted. As I see
that some who have been involved are not on this distribution list. 

Bye for now,
George Penokie

Dept 2C6  114-2 N212
E-Mail:    gop at us.ibm.com
Internal:  553-5208
External: 507-253-5208   FAX: 507-253-2880






Jim Hafner/Almaden/IBM 

09/29/2004 04:30 PM 


To

George Penokie/Rochester/IBM at IBMUS 


cc

"Mallikarjun C." <cbm at rose.hp.com>, "Reuter, Jim" <james.reuter at hp.com>,
Julian Satran <Julian_Satran at il.ibm.com>, "KRUEGER,MARJORIE
(HP-Roseville,ex1)" <marjorie.krueger at hp.com>, Paul.Vonbehren at sun.com,
T10 Reflector <t10 at t10.org> 


Subject

Re: Can one LU be in multiple devices? Link
<Notes://062564740060D5BE/A24CF79FA927570285256356005827C4/E59B7D20F4A18
0E386256F1E006FE221> 

 


 

 




George, 

What I read now is that there is a constraint that one LU per domain,
but that doesn't preclude an LU from appearing in multiple domains.  Is
that right? 

But if so, what is the point of a restriction about one LU (uniquely
named) per domain.  What feature is gained by this restriction?   

Consider the following two entities: 
1) dual-headed controller with shared LUs that collectively form a
single target device (one name) 
2) dual-headed controller with shared LUs that form a pair of target
devices. 

In what you say, the second is outlawed, UNLESS they are on separate
domains (e.g., FC loops), in which case it's OK.   But what's the
difference functionally? Why does the domain pose any constraints on the
model? 

Thanks, (and I've never seen you so verbose before :-) 

Jim Hafner/Almaden/IBM
Tel: (408) 927-1892, Fax: (408) 927-3030 (t/l 457)
Email:hafner at almaden.ibm.com 




 

George Penokie 

09/29/2004 02:15 PM 


        To:        Jim Hafner/Almaden/IBM at IBMUS 
        cc:        "Mallikarjun C." <cbm at rose.hp.com>, "Reuter, Jim"
<james.reuter at hp.com>, Julian Satran <Julian_Satran at il.ibm.com>,
"KRUEGER,MARJORIE (HP-Roseville,ex1)" <marjorie.krueger at hp.com>,
Paul.Vonbehren at sun.com, T10 Reflector <t10 at t10.org> 
        Subject:        Re: Can one LU be in multiple devices? Link
<Notes://882563A90059E2B4/5E502A1BAAAF40CA85256197006C1A32/C38CF0850CB05
C0C85256F1E006E1049> 



Jim, 

In SCSI there are names and identifiers. Identifiers are unique to the
SCSI domain. Names are required to be world wide unique (i.e., you
should only see the name of a SCSI device once no matter what). 

So the logical units name is worldwide unique therefore it doesn't
matter where it is, who is talking to, whether it is virtual, or real it
has to have a unique name. Now it can have more that one unique name but
that same set of names are always returned from that logical unit when
request regardless of which protocol asks for them. There is no rule
that prevents a single logical unit from appearing in more than one SCSI
domain. In fact, the name is how a management applications can tell it's
the same logical unit even though it is being accessed from different
SCSI domains. 

An underlaying physical logical unit will have one (set) unique name(s)
and any virtual (mapped) logical unit must have a different (set) of
unique name(s). 

You certainly like those SCSI target devices. But from the point of view
of how they look to different domains and the naming rules there is
really no difference between logical units and SCSI target devices.
Except it's a heck of a lot easier to get the logical units name. But if
you insist: 

A SCSI target devices name is worldwide unique therefore it doesn't
matter where it is, who is talking to, whether it is virtual, or real it
has to have a unique name. Now it can have more that one unique name but
that same set of names are always returned regardless of which protocol
asks for them. There is no rule that prevents a single SCSI target
device from appearing in more than one SCSI domain. 

In case you are wondering there is two ways to find out a SCSI target
devices name in SCSI but only one is guaranteed to work. You can do an
INQUIRY command and asked for VPD page 83h and hope the logical unit
sends you the SCSI target devices name. Then you can do that for every
logical unit in you know about and compare the SCSI target device names
(assuming they all return that information). Or you can send the INQUIRY
command and a REPORT LUNS command to a REPORT LUNS well known logical
unit and it is required to tell the SCSI target devices name and give
you a list of all the logical units in that SCSI target device. 

So now I suspect your next question will be: But if everything you say
is true can a single logical unit appear in two different SCSI target
devices if those SCSI target devices are in two different SCSI domains. 

The answer is yes and long as the SCSI domain rules are followed.
Because they are in different SCSI domains and SCSI rules currently only
apply to the SCSI domain. 

Bye for now,
George Penokie

Dept 2C6  114-2 N212
E-Mail:    gop at us.ibm.com
Internal:  553-5208
External: 507-253-5208   FAX: 507-253-2880






Jim Hafner/Almaden/IBM 

09/29/2004 03:09 PM 


To

George Penokie/Rochester/IBM at IBMUS 


cc

"Mallikarjun C." <cbm at rose.hp.com>, "Reuter, Jim" <james.reuter at hp.com>,
Julian Satran <Julian_Satran at il.ibm.com>, "KRUEGER,MARJORIE
(HP-Roseville,ex1)" <marjorie.krueger at hp.com>, Paul.Vonbehren at sun.com,
T10 Reflector <t10 at t10.org> 


Subject

Re: Can one LU be in multiple devices? Link
<Notes://062564740060D5BE/A24CF79FA927570285256356005827C4/5CFB9927A337B
5EC86256F1E0064D86C> 

 


 

 



George, 

Thanks for the history lesson.  But I think you confused me. 

You state: 
The net of all of this is that: 
- a single logical unit in one SCSI domain may be mapped into several
logical units in another domain. 
- a single logical unit in a single target device in one SCSI domain may
be mapped into several logical units with any number of those logical
units residing on any number of target devices in another domain. 

BUT the rules for naming still apply. 
-All the logical unit names have to be worldwide unique. 
-All the target device names have to be worldwide unique. 

So, if a single logical unit (call it the source) is "mapped into
several logical units", and each logical unit name is WWU, does this
mean that the "mapped" LUs have to have a different name from the source
one?  If not, then it would appear that the same LU shows up in multiple
target devices.  If so, then there is no (standard) way from a
management perspective to tell that the mapped LUs have a source. 

Also, you seem to imply (if I read you right) that in order for an LU to
appear in another domain, it must be in some different target device.
This presumes then that a target device lives in only one domain -- but
can't a device have both FC, SPI and iSCSI interface and so appear in
multiple domains (or even two SPI interfaces on different SCSI buses)?
Or is this a case of "two devices" with the same target device name? 

Thanks, 

Jim Hafner/Almaden/IBM
Tel: (408) 927-1892, Fax: (408) 927-3030 (t/l 457)
Email:hafner at almaden.ibm.com 




 

George Penokie 

09/29/2004 12:58 PM 


        To:        Jim Hafner/Almaden/IBM at IBMUS 
        cc:        "Mallikarjun C." <cbm at rose.hp.com>, "Reuter, Jim"
<james.reuter at hp.com>, Julian Satran <Julian_Satran at il.ibm.com>,
"KRUEGER,MARJORIE (HP-Roseville,ex1)" <marjorie.krueger at hp.com>,
owner-t10 at t10.org, Paul.Vonbehren at sun.com, T10 Reflector <t10 at t10.org> 
        Subject:        Re: Can one LU be in multiple devices? Link
<Notes://882563A90059E2B4/5E502A1BAAAF40CA85256197006C1A32/7F7FF38D60E51
CC285256F1E005059B9> 



I have been reading these notes, looking in the places that supposedly
justify the belief that a logical unit can be split across multiple
target devices. But so far all I have found is that, either target
devices are not talked about or that target is being used, either
directly or indirectly, to refer to a target port. 

So a little history lesson is in order. Up until a year or so the term
target was a fuzzy thing on one end of a wire (with an initiator on the
other). This fuzziness started to become a problem when the SCSI
architecture began expanding outside the parallel environment. To
resolve this the SCSI committee started to look into defining the
various SCSI objects (soon to become classes) more solidly. The first,
and most obvious, problem was that 90% of the time the term target or
initiator was used in the standards what was really being described was
the port on a target or initiator. As a result new terms were defined:
target port, target device, initiator port, and initiator device.  This
has resulted in every use of the term target and initiator to being
examined to determine if it is a port, a device, or something else. At
this point SAM-3, SBC-2, SPC-3, and SAS 1.1 have been examined and
corrected to the new terminology. 

Back to the main point. Under SAM-3 a target device contains logical
units, target ports, and has a name (it has other objects but that is a
different discussion). There is no provision in SAM-3 that would allow a
single logical unit to be split across multiple target devices and there
is no amount of SAM-3 word twisting that will change that. This is not
something that I believe is desirable or necessary to accomplish given
everything that I have read so far.   

Now, before you vitalization guys go crazy, keep reading. 

Vitalization of logical units has been implemented in SCSI long before
that term was in vogue. It worked under SCSI then and it still works.
The reason it works is because of how the SCSI domain works. For example
a RAID device or a bridge controller that receives SCSI operations
(i.e., a target device) and then retransmits the information received
using different SCSI operations (i.e., an initiator device) operates in
two SCSI domains. Each SCSI domain has a set of initiator ports, target
ports, logical units, and, if you insist, target devices. The
relationship between the all these objects within each domain is
strictly controlled by the SCSI standards. The relationship between the
objects between the SCSI domains is not currently defined by SCSI. 

The net of all of this is that: 
- a single logical unit in one SCSI domain may be mapped into several
logical units in another domain. 
- a single logical unit in a single target device in one SCSI domain may
be mapped into several logical units with any number of those logical
units residing on any number of target devices in another domain. 

BUT the rules for naming still apply. 
-All the logical unit names have to be worldwide unique. 
-All the target device names have to be worldwide unique. 

As a final note: SAM-3 is done and SAM-4 is starting. The most obvious
change in SAM-4 will be the use of UML diagrams. Another change is the
editor has is now myself. So if there is some new class that needs to be
defined (as Paul suggests) then someone will have to bring in a proposal
and get it accepted. 

Bye for now,
George Penokie

Dept 2C6  114-2 N212
E-Mail:    gop at us.ibm.com
Internal:  553-5208
External: 507-253-5208   FAX: 507-253-2880

<<<Filtered>>> 







------=_NextPart_000_002E_01C4AAA6.E53C8BE0
Content-Type: text/html;
	charset="us-ascii"
Content-Transfer-Encoding: quoted-printable


 
From: owner-t10 at t10.org [mailto:owner-t10 at t10.org] On Behalf Of Jim Hafner
 Sent: Monday, October = 04, 2004 4:23 PM
 To: George Penokie
 Cc: Mallikarjun C.; = Reuter, Jim; Julian Satran; KRUEGER,MARJORIE (HP-Roseville,ex1); = Paul.Vonbehren at sun.com; T10 Reflector
 Subject: Re: Can one LU = be in multiple devices? 
  
George, 

In the first place, I'd never try to convince you that you are = ;wrong; :-).   

And I believe that what you say is the correct interpretation of the = current SAM-3.   

And I have no intention of trying to change this even for SAM-4 (I have no = vested interest in this -- at least not at this time).  I'm really trying = to help out Mallikarjun and friends who brought the question to me (and I = apparently answered wrong!). 

But, I'm not sure you answered my last questions.  The key to me is = this one: 
** Why does the domain impose this LU constraint on the model? ** = 
In other words, what inherent property of the domain requires the = constraint for model consistency? (Or is this just an historical fact?). =   
Or from a different viewpoint, what feature does this constraint provide = within a domain?  And how is this feature handled across domains? (Or is it = simply that the standard only addresses ;within a domain; and = doesn't address ;cross-domain; issues?) 

Thanks, 

Jim Hafner/Almaden/IBM
 Tel: (408) 927-1892, Fax: (408) 927-3030 (t/l 457)
 Email:hafner at almaden.ibm.com 

  George = Penokie 09/30/2004 01:15 PM 
        To:     =    Jim Hafner/Almaden/IBM at IBMUS 
        cc:        ;Mallikarjun C.; <cbm at rose.hp.com>, ;Reuter, Jim; <james.reuter at hp.com>, Julian Satran = <Julian_Satran at il.ibm.com>, ;KRUEGER,MARJORIE (HP-Roseville,ex1); <marjorie.krueger at hp.com>, Paul.Vonbehren at sun.com, T10 = Reflector <t10 at t10.org> 
        Subject:        Re: Can one = LU be in multiple devices? 

Jim, 

The constraint is that one logical unit cannot span multiple target devices = when both those targets devices and the logical unit are in the same domain. = (This is a very simplified statement that is described in much more gory = detail in my previous note. 

I believe your example and the statement after is correct. = 

You ask why it is this way. My question is what's wrong with the current definition. As far as I know, that are now cases where it causes a = problem. Also, allowing it would cause problems in identifying and locating = specific target devices. 

Before you start trying to convince my I am wrong, STOP, it will do no good. = The standard is the way I have described it and SAM-3 is done. To change SAM-4 you = have to convince the CAP working group, not me,  not only that a change is = needed but you will have to write up all the changes to the standards that = would be required to implement the change. 

Consider these notes an education as to how it is defined today and go from = there, if you dare. 

Bye the way the t10 reflector has been dead for most of this thread. When = it comes up I will try to make sure this is all posted. As I see that some who = have been involved are not on this distribution list. 

Bye for now,
 George Penokie
 
Dept 2C6  114-2 N212
 E-Mail:    gop at us.ibm.com
 Internal:  553-5208
 External: 507-253-5208   FAX: 507-253-2880
 


Jim = Hafner/Almaden/IBM 09/29/2004 04:30 PM To George = Penokie/Rochester/IBM at IBMUS cc ;Mallikarjun C.; <cbm at rose.hp.com>, ;Reuter, Jim; <james.reuter at hp.com>, Julian Satran <Julian_Satran at il.ibm.com>, ;KRUEGER,MARJORIE (HP-Roseville,ex1); <marjorie.krueger at hp.com>, = Paul.Vonbehren at sun.com, T10 Reflector <t10 at t10.org> Subject Re: Can one LU be in multiple = devices?       


George, 

What I read now is that there is a constraint that one LU per domain, but = that doesn't preclude an LU from appearing in multiple domains.  Is = that right? 

But if so, what is the point of a restriction about one LU (uniquely named) = per domain.  What feature is gained by this restriction?   = 

Consider the following two entities: 
1) dual-headed controller with shared LUs that collectively form a single = target device (one name) 
2) dual-headed controller with shared LUs that form a pair of target = devices. 

In what you say, the second is outlawed, UNLESS they are on separate = domains (e.g., FC loops), in which case it's OK.   But what's the = difference functionally? Why does the domain pose any constraints on the model? = 

Thanks, (and I've never seen you so verbose before :-) 

Jim Hafner/Almaden/IBM
 Tel: (408) 927-1892, Fax: (408) 927-3030 (t/l 457)
 Email:hafner at almaden.ibm.com 

  George = Penokie 09/29/2004 02:15 PM 
        To:     =    Jim Hafner/Almaden/IBM at IBMUS 
        cc:        ;Mallikarjun C.; <cbm at rose.hp.com>, ;Reuter, Jim; <james.reuter at hp.com>, Julian Satran = <Julian_Satran at il.ibm.com>, ;KRUEGER,MARJORIE (HP-Roseville,ex1); <marjorie.krueger at hp.com>, Paul.Vonbehren at sun.com, T10 = Reflector <t10 at t10.org> 
        Subject:        Re: Can one = LU be in multiple devices? 

Jim, 

In SCSI there are names and identifiers. Identifiers are unique to the = SCSI domain. Names are required to be world wide unique (i.e., you should = only see the name of a SCSI device once no matter what). 

So the logical units name is worldwide unique therefore it doesn't matter = where it is, who is talking to, whether it is virtual, or real it has to have a = unique name. Now it can have more that one unique name but that same set of = names are always returned from that logical unit when request regardless of which protocol asks for them. There is no rule that prevents a single logical = unit from appearing in more than one SCSI domain. In fact, the name is how a management applications can tell it's the same logical unit even though = it is being accessed from different SCSI domains. 

An underlaying physical logical unit will have one (set) unique name(s) = and any virtual (mapped) logical unit must have a different (set) of unique = name(s). 

You certainly like those SCSI target devices. But from the point of view of = how they look to different domains and the naming rules there is really no difference between logical units and SCSI target devices. Except it's a = heck of a lot easier to get the logical units name. But if you = insist: 

A SCSI target devices name is worldwide unique therefore it doesn't = matter where it is, who is talking to, whether it is virtual, or real it has to have = a unique name. Now it can have more that one unique name but that same = set of names are always returned regardless of which protocol asks for them. = There is no rule that prevents a single SCSI target device from appearing in = more than one SCSI domain. 

In case you are wondering there is two ways to find out a SCSI target = devices name in SCSI but only one is guaranteed to work. You can do an INQUIRY = command and asked for VPD page 83h and hope the logical unit sends you the SCSI = target devices name. Then you can do that for every logical unit in you know = about and compare the SCSI target device names (assuming they all return that information). Or you can send the INQUIRY command and a REPORT LUNS = command to a REPORT LUNS well known logical unit and it is required to tell the = SCSI target devices name and give you a list of all the logical units in = that SCSI target device. 

So now I suspect your next question will be: But if everything you say is = true can a single logical unit appear in two different SCSI target devices if = those SCSI target devices are in two different SCSI domains. 

The answer is yes and long as the SCSI domain rules are followed. Because = they are in different SCSI domains and SCSI rules currently only apply to the = SCSI domain. 

Bye for now,
 George Penokie
 
Dept 2C6  114-2 N212
 E-Mail:    gop at us.ibm.com
 Internal:  553-5208
 External: 507-253-5208   FAX: 507-253-2880
 


Jim = Hafner/Almaden/IBM 09/29/2004 03:09 PM To George = Penokie/Rochester/IBM at IBMUS cc ;Mallikarjun C.; = <cbm at rose.hp.com>, ;Reuter, Jim; <james.reuter at hp.com>, Julian Satran <Julian_Satran at il.ibm.com>, ;KRUEGER,MARJORIE (HP-Roseville,ex1); <marjorie.krueger at hp.com>, Paul.Vonbehren at sun.com, T10 Reflector = <t10 at t10.org> Subject Re: Can one LU be in multiple = devices?       

George, 

Thanks for the history lesson.  But I think you confused me. = 

You state: 
The net of all of this is that: 
- a single logical unit in one SCSI domain may be mapped into several = logical units in another domain. 
- a single logical unit in a single target device in one SCSI domain may = be mapped into several logical units with any number of those logical = units residing on any number of target devices in another domain. 

BUT the rules for naming still apply. 
-All the logical unit names have to be worldwide unique. 
-All the target device names have to be worldwide unique. 

So, if a single logical unit (call it the source) is ;mapped into = several logical units;, and each logical unit name is WWU, does this mean = that the ;mapped; LUs have to have a different name from the source = one?  If not, then it would appear that the same LU shows up in multiple target = devices.  If so, then there is no (standard) way from a management = perspective to tell that the mapped LUs have a source. 

Also, you seem to imply (if I read you right) that in order for an LU to = appear in another domain, it must be in some different target device.  This = presumes then that a target device lives in only one domain -- but can't a = device have both FC, SPI and iSCSI interface and so appear in multiple domains (or = even two SPI interfaces on different SCSI buses)?   Or is this a case of = ;two devices; with the same target device name? 

Thanks, 

Jim Hafner/Almaden/IBM
 Tel: (408) 927-1892, Fax: (408) 927-3030 (t/l 457)
 Email:hafner at almaden.ibm.com 

  George = Penokie 09/29/2004 12:58 PM 
        To:     =    Jim Hafner/Almaden/IBM at IBMUS 
        cc:        ;Mallikarjun C.; <cbm at rose.hp.com>, ;Reuter, Jim; <james.reuter at hp.com>, Julian Satran = <Julian_Satran at il.ibm.com>, ;KRUEGER,MARJORIE (HP-Roseville,ex1); <marjorie.krueger at hp.com>, owner-t10 at t10.org, = Paul.Vonbehren at sun.com, T10 Reflector <t10 at t10.org> 
        Subject:        Re: Can one = LU be in multiple devices? 

I have been reading these notes, looking in the = places that supposedly justify the belief that a logical unit can be split = across multiple target devices. But so far all I have found is that, either = target devices are not talked about or that target is being used, either = directly or indirectly, to refer to a target port. 

So a little history lesson is in order. Up until a year or so the term = target was a fuzzy thing on one end of a wire (with an initiator on the other). = This fuzziness started to become a problem when the SCSI architecture began = expanding outside the parallel environment. To resolve this the SCSI committee = started to look into defining the various SCSI objects (soon to become classes) = more solidly. The first, and most obvious, problem was that 90% of the time = the term target or initiator was used in the standards what was really being = described was the port on a target or initiator. As a result new terms were = defined: target port, target device, initiator port, and initiator device. =  This has resulted in every use of the term target and initiator to being = examined to determine if it is a port, a device, or something else. At this point = SAM-3, SBC-2, SPC-3, and SAS 1.1 have been examined and corrected to the new terminology. 

Back to the main point. Under SAM-3 a target device contains logical units, = target ports, and has a name (it has other objects but that is a different discussion). There is no provision in SAM-3 that would allow a single = logical unit to be split across multiple target devices and there is no amount = of SAM-3 word twisting that will change that. This is not something that I = believe is desirable or necessary to accomplish given everything that I have read = so far.   

Now, before you vitalization guys go crazy, keep reading. 

Vitalization of logical units has been implemented in SCSI long before that term was = in vogue. It worked under SCSI then and it still works. The reason it = works is because of how the SCSI domain works. For example a RAID device or a = bridge controller that receives SCSI operations (i.e., a target device) and = then retransmits the information received using different SCSI operations = (i.e., an initiator device) operates in two SCSI domains. Each SCSI domain has a = set of initiator ports, target ports, logical units, and, if you insist, = target devices. The relationship between the all these objects within each = domain is strictly controlled by the SCSI standards. The relationship between the = objects between the SCSI domains is not currently defined by SCSI. = 

The net of all of this is that: 
- a single logical unit in one SCSI domain may be mapped into several = logical units in another domain. 
- a single logical unit in a single target device in one SCSI domain may = be mapped into several logical units with any number of those logical = units residing on any number of target devices in another domain. = 

BUT the rules for naming still apply. 
-All the logical unit names have to be worldwide unique. 
-All the target device names have to be worldwide unique. 

As a final note: SAM-3 is done and SAM-4 is starting. The most obvious = change in SAM-4 will be the use of UML diagrams. Another change is the editor = has is now myself. So if there is some new class that needs to be defined (as = Paul suggests) then someone will have to bring in a proposal and get it = accepted. 

Bye for now,
 George Penokie
 
Dept 2C6  114-2 N212
 E-Mail:    gop at us.ibm.com
 Internal:  553-5208
 External: 507-253-5208   FAX: 507-253-2880
 
<<<Filtered>>> 







------=_NextPart_000_002E_01C4AAA6.E53C8BE0--




More information about the T10 mailing list