SAM Revision 14

Charles Monia, SHR3-2/W3, 237-6757 11-Aug-1994 1633 monia at starch.enet.dec.com
Thu Aug 11 13:31:43 PDT 1994


Gary Stephens' note says in part:

>Clause 3.2.1, para. 2, page 25 says that an application client is created
>for each COMMAND.  This is inconsistent with SCSI-2 CAM which creates one
>CCB for each command in a linked list of commands.  That is, an
>application client is created for each task.

The notion of having one application client per task  versus creating a new
client for each  command was debated extensively at the SAM working group held
the Sunday before the January 1992 plenary. After arguing the issue 
unsuccessfully at that time , the technical editor decided it was better
to switch than fight and reluctantly acquiesced to the present model.

I still feel the notion in SAM is counterintuitive. One thinks of a series
of linked commands as representing an extended I/O operation of some sort.
which is under the control of a single "intelligence" in the initiator. Defining
a model that requires creation of a seperate thread of execution (ie.
application client) for each CDB in the sequence adds complexity without adding 
value.

Although the present model is functionally adequate and correct, it doesn't map 
very well onto the way host applications are built -- as CAM illustrates. In any 
event, I believe this is a matter of aesthetics rather than a functional flaw in 
the model and can be left as it is.

Charles







More information about the T10 mailing list