Summary and comparison of alternative multiple port management proposals

Charles Monia, SHR3-2/W3, 237-6757 23-Nov-1994 1507 monia at starch.enet.dec.com
Wed Nov 23 12:16:46 PST 1994


From:	US4RMC::"Bob.Snively at eng.sun.com" "Bob Snively" 23-NOV-1994 01:10:28.37
To:	scsi at WichitaKS.ATTGIS.COM
CC:	
Subj:	Summary and comparison of alternative multiple port management proposals

I support the second alternative, with the following suggested modifications


>2)	MULTIPLE PORT IMPLEMENTATION

>The present revision of document X3T10/94-233 defines two functions
>required for multiple port operation.  The number of ports is not
>limited. The mechanism for defining the ports is outside the scope
>of the standards, but the following methods may be used.
>
>	Targets can be identified by their serial number or, if
>	supported by the SCSI protocol, their world-wide name.
>	Initiators communicate by methods outside the protocol to
>	compare

(The last sentence seems to have gotten truncated.)

The methods you propose to seem to define ways of
identifying the target independant of the port through which
the target is addressed. At least I hope so.

If that's the case, it seems to me that the main thrust of
this proposal is multi-initiator, not multi-port configurations.


I>	Initiators can be identifed by their world-wide name.

I think it would simplify matters to merge the definition
of  target and initiator identifiers into a single 'global device
identifier',  which is unique within the scope of the scsi domain. 
The manner in which global device identifiers are established
would be protocol-specific, within the following constraints:

1. Each SCSI device shall have at least one global device identifier
   (GDI).

2. If a devine has more than one GDI, it appears to the
    system as multiple physical devices, one per GDI.

Although protocols such as SSP and SBP may provide
for other protocol-specific identifiers such as hop count
or login identifier, the GDI would be the only device
identifier used in all task management functions and SCSI
commands.

>	For SIP devices which have only two ports, targets identify
>	initiators as being  those installed on this port or those
>	installed on the other port.

IMO: The previosly defined dual-port function in SIP should be treated as an
ad-hoc special case, which should not be extended into the generalized
multi-port architecture you've proposed.

Thanks,
Charles





More information about the T10 mailing list