CRN_quandary

Dal Allan endlcom at ibm.net
Mon Jul 6 19:32:59 PDT 1998


* From the T10 (formerly SCSI) Reflector (t10 at symbios.com), posted by:
* "Dal Allan" <endlcom at ibm.net>
*
          Subject:  CRN

In Florida I believe the long discussion on CRN settled on the following. 

 - CRN capability is established at PRLI for the target and affects all 
LUNs.
  - CRNs are maintained by LUN, because the in-order nature is meaningful to 
the LUN, not the target. 
 - Task management commands cannot have a CRN otherwise they could never get 
through if a delivery error had occurred. 
 - The CRN is restored to a known state by Target Reset or another PRLI. 

Based on the above, the CRN is being treated as both LUN-specific and 
target-generic. 

On a RAID controller with a bunch of disk drives that has a tape drive as 
well, a Target Reset or PRLI will blow off all the queued disk drive 
commands because of a delivery error to the tape drive. 

The ordering error that affected the tape drive has to be recovered, but why 
should that affect every other LUN on the target? 

If CRN is LUN-specific then that should mean:

 - Device Reset 
 - Mode Page defines support 
 - CRN counters for each LUN 

If CRN is target generic then that should mean:

 - Target Reset 
 - PRLI defines support 
 - One CRN counter for the target 
 - The target is responsible for ordered delivery to LUNs that are targets 
in their own right e.g. 

            +------------------------------------------+
            |                                          |
  Initiator +----------------Host----------------------+
                               | 
                               |   
                               | 
  Target    +----------------RAID----------------------+  Initiator
            |                                          |    
            +---+----+----+----+-----+-------------+---+  
  LUN          Disk Disk Disk Disk Disk Disk      Tape     Target

The above is grossly simplified because what the host thinks are LUNs and 
what the RAID controller portrays them to be are a confusion factor. Look on 
the above as a JBOD to avoid perturbations of logical and real devices. 

The tape working group needs to decide whether it wants CRN to be a target 
behavior or a LUN behavior because CRN is now neither one nor the other. 

There needs to be architectural consistency on the way CRN is treated.


Dal
*
* 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