Re: Fixing SCSI Layer

[Tony Chung]

Gerard Roudier wrote:

 

UDI is not a standard and seem to me some poor sci-fi and will never
happen in my systems.

I just browse http://www.projectudi.org/ and find out that Intel already
porting UDI  in Linux as proof of concept.
 

CAM requires the data direction (or none) to be provided in the Execute
SCSI IO request CCB, but existing Linux applications donnot provide the
data direction since the current SCSI interface just does not allow this
information to be passed. We must deal with compatibility issues.

  Current linux-scsi:
                    (High level drivers eg. sd, st)  <--> (middle layer)  <---> ( host adaptor layer)

Future layers:
 
                  (Old drivers)  <---->  (middle layer *)  <---(UDI SCSI  layer) <---> (old host adaptor layer)
                                                                         ^--------------------------------^
                   (New  perihperal drivers)  <------------- >(UDI SCSI layer) <---> (new  UDI host adaptor drivers)

With multiple paths, you can maintain backward binary compatible and user still can
pick the fastest path for efficiency.
If old drivers do not pass DATA_DIRECTION, then the middle layer can figure it out and pass it
to the UDI SCSI layer and take advantages of latest hardware.  If not, they have to write their own
perihperal drivers.

 
 

The better one is the one that is a real standard. The other one is to be
ignored, in my opinion.

You may want to look at UDI more carefully because:
1. support by major vendors like Sun, HP, IBM and Compaq (which buyout DEC and abandon CAM).
2. Intel is porting UDI for Linux now!
3. UDI support SCSI-3 (including Fibre Channel)...Do CAM support that?

 

Gérard.

-- 
=============================
Tony Chung