RE: SCSI URL scheme [WAS: Re: iSCSI: 2.2.6. Naming & mapping]

[Howard Green <hgreen@Brocade.COM>]

Sorry for the previous incomplete post. My mailer misfired...

I know that the following is kind of a radical question to ask, but has the time arrived to toss out LUNs as a means of identifying logical units, in favor of simply using the logical units' WWNs directly?

Back in the good ol' days, LUNs were small integers contained directly in each CDB, and were (in practice, if not in theory) a simple index assigned to a specific logical unit throughout its life. SCSI has obviously evolved to a far more sophisticated state of affairs. The LUNs used to identify a logical unit have grown to 64 bits, and can vary over time, vary relative to the target interface used, and vary as a function of the initiator's identity. Clearly, they're not durable identifiers in any useful sense. Further, they've got to be be set up and administered. My sense here is that they've become just so much bureaucracy, to no particularly useful end.

Currently, an IU containing a SCSI command identifies the target logical unit by a Logical Unit Number. So, in effect, the notion here would be to use the logical unit's WWN in place of the LUN in the command IU. The WWNs I've seen used to identify logical units currently range from 64 to 128 bits, so there's no particular additional overhead involved. 

I think I'm reasonably cognizant of the difficulties imposed by making such a change (changes to SAM, compatibility with existing implementations, etc.)---and that it's really a T10 issue, and not exclusively an iSCSI issue. However, my sense is that the current LUN-based identification mechanism is causing increasing grief, as evidenced by any number of recent postings; and one has to start somewhere...

Howard Green


> -----Original Message-----
> From: Raghavendra Rao [mailto:jpr@divyaroot.India.Sun.COM]
> Sent: Tuesday, October 03, 2000 10:56 AM
> To: ips@ece.cmu.edu
> Subject: Re: SCSI URL scheme [WAS: Re: iSCSI: 2.2.6. Naming & mapping]
> 
> 
> 
> > out-of-band).  I don't see the point of the initiator 
> sending its "view" to
> > the target at any point.  If the initiator sends an 
> incorrect view, you've
> > just created an error scenario which needs additional definition.
> 
> Well, to extend what Julian said further, the initiator could 
> ask the target
> for a LUN value corresponding to a WWN as part of the 
> authentication process,
> instead of presenting its view - whether right or wrong. But 
> something like
> this needs to be done as there is no easier/faster way to 
> translate a LU WWN
> to a LUN value.
> 
> This may also help in building an authentication method for a 
> LU access where
> no such thing exists today ?
> 
> 
> > case, the target will send its "view" for that initiator in 
> REPORT LUNS
> > SCSI command after login (which is actually the layer that 
> cares about the
> > LUN Map at all anyway).
> > 
> 
> This will be a problem if the storage controller has a lot of 
> LUNs. Since
> REPORT LUNS only returns a list of LUNs, the initiator still 
> has to send an
> INQUIRY (page 83) request to each LU to to find a matching LU 
> WWN (or until
> a matching WWN is found).
> 
> -JP
>