Definition of mandatory

Charles Monia, SHR3-2/W3, 237-6757 08-Sep-1994 1732 monia at starch.enet.dec.com
Thu Sep 8 14:30:30 PDT 1994


Quoting from Jim McGrath's note:

>I agree that the whole task set/queuing sections are unique to SAM.  My
>problem is that, as they are currently written, they only provide an
>architecture for the management of queuing.  Specifically, they leave the
>location (host, target) of the management of the various elements of the task
>set largely undefined. 

I'd go further and say that SAM intentionally leaves it to the protocol standard 
to define how architectural entities (logical units, tasks sets, etc) map
to an implementation.

>..........................That is, it is perfectly SAM compliance to, say,
>implement ordered tasks by having a device driver withhold the issuing of
>ordered commands until the device is idle (no queue entries), and then send
>them one at a time.  In this case, the target need never see an ordered tag. 
>Indeed, I am counting on this interpretation to help make my proposal on ATA
>command queuing more "SAM compliant", even though it will only require simple
>tags.

A protocol standard claiming to conform to SAM must show compliance at the 
intiator's protocol service interface. In my opinion, therefore, one could
write a compliant protocol standard that was based on the scenario you describe.


>Given this, I cannot approve SAM if the interpretation is that it applies to
>devices, 

I believe the extent to which SAM applies to some physical device is determined
by the applicable protocol standard. For example, there is a very close mapping 
between SIP's notion of a target device and the target device model described in 
SAM. Any logical unit functions defined in SAM would therefore be implemented 
within a SIP device and would have to behave according to the rules specified in 
SAM.

In the ATA variant you mention, the functions of the logical unit are 
distributed between the host and the ATA device. If the protocol standard 
delegates a specific SAM function to an ATA device, such as providing SCSI
status for a completed command, then that function must behave as specified by
SAM. e.g., a CHECK CONDITION status returned by the ATA device must have the 
encoding specified by SAM.

>........for in the interests of defining a general architecture (a good and
>proper thing to do), it fails to give enough direction to design a compliant
>implementation at a target level,

I think we agree that the job of the protocol and interconnect standards is to 
provide such direction.


Charles










More information about the T10 mailing list