FCP-2: Command reference per LUN????
matt_wakeley at agilent.com
Tue Jun 27 14:00:10 PDT 2000
* From the T10 Reflector (t10 at t10.org), posted by:
* Matt Wakeley <matt_wakeley at agilent.com>
"David A. Peterson" wrote:
> The issue of whether CRN is target or lun based was discussed quite a
> while back and I argued the same points as Matt below but lost the
> battle. Given that, I persued the concept of adding the CRN to the SAM-2
> model with no luck (at the time).
> The main (only?) reason CRN was made lun based was that the disk drive
> target/lun did not want to see the CRN in the FCP command payload.
I don't understand the above argument when you say "target/lun". The current
definition is "LU" handles the CRN. I'm saying the Target should handle it.
Disk drives that don't care about ordering can just ignore the CRN field.
> still believe the use of a continuously increasing OX_ID can acheive the
> same result as a target based CRN (i.e it's unfortunate we need to have
> the concept of CRN is the first place).
> Matt Wakeley wrote:
> > *
> > * From the fc reflector, posted by:
> > * Matt Wakeley <matt_wakeley at agilent.com>
> > *
> > In FCP-2, it says in section 4.3:
> > "Precise delivery of SCSI commands uses the COMMAND REFERENCE NUMBER (CRN) in
> > the FCP_CMND IU. For each device server having the EPDC bit set to one, the
> > application client places a monotonically increasing one byte integer in the
> > CRN field for each command that is transmitted that also requires precise
> > delivery."
> > Where "device server" is defined in 3.1.15 as "An object within the logical
> > unit which executes SCSI tasks and enforces the rules for task management."
> > And in 184.108.40.206 it says "The COMMAND REFERENCE NUMBER (CRN) is provided by the
> > initiator to assist in performing precise delivery checking for FCP commands.
> > If precise delivery is enabled, a nonzero value of CRN shall be treated as a
> > command reference number in determining the receipt and ordering of commands
> > from a particular initiator to the particular logical unit as described in
> > 4.3."
> > All this says that the CRN is maintained on a per LU basis. This will require
> > that FCP maintain a "table" of CRNs for every LU in every SCSI target that it
> > is communicating with. Remember, SCSI does not have any concept of CRN.
> > I thought the understanding during the discussions of this, that that CRN was
> > based on a per TARGET, *not* LU behind the target.
> > In other words, the SCSI driver passes commands to the FCP layer to be
> > delivered across fibre channel to the target. SCSI does not have a concept of
> > CRN, so, no CRN is passed to the FCP driver. FCP has to guarantee that these
> > commands are delivered back to SCSI at the target in the same order that the
> > SCSI driver passed the commands to FCP (as per SAM-2).
> > So, what these words are requiring is that SCSI will have to be modified to
> > track this CRN on a per LU basis. FCP-2 does not have the knowledge of
> > information of what's stored in LU control pages (the EPDC bit). If FCP is
> > required to have knowledge of all the LUs behind all the targets it is
> > communicating to, this is violating the boundary between what is FCP and what
> > is SCSI.
> > Also, if FCP is required to keep track of all the possible CRNs for each LU
> > behind each target that it is communicating with, that is a lot of state to
> > keep around in chips and/or firmware and/or fibre channel drivers.
> > FCP(-2) is a "mapping" of SCSI to be delivered across Fibre Channel. That is,
> > it is a Transport across FC. It should not have to have knowledge about the
> > internals of SCSI (details about mode pages, number of LUs behind a target,
> > etc).
> > In SAM-2, the SCSI Command Model is defined as:
> > Service response =Execute Command (Task Address, CDB, [Task Attribute],
> > [Data-Out Buffer], [Command Byte Count], [Autosense Request] || [Data-In
> > Buffer], [Sense Data], Status)
> > No where is there a CRN. Is FCP-2 placing a new requirement on SCSI (SAM-2) to
> > include CRN's in it's command model?
> > The text in FCP-2 needs to be changed to eliminate this notion of CRNs per
> > LU. The CRN was meant to be on a per Target basis.
> > -Matt Wakeley
> > Agilent Technologies
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org
More information about the T10