de facto obsolete sbc-2 ops - x 8B 9E 9F

Pat LaVarre LAVARRE at iomega.com
Thu Feb 21 07:05:51 PST 2002


* From the T10 Reflector (t10 at t10.org), posted by:
* "Pat LaVarre" <LAVARRE at iomega.com>
*
>>> <Gerry.Houlder at seagate.com> 02/20/02 11:42AM >>>
> Actually we are trying to obsolete
> the entire idea of a SEEK command.
> ..There was an attempt to obsolete
> these commands that was nearly passed.

Ah.  Anybody know something of the date/Url of that proposal?  The search <http://www.google.com/search?as_q=seek&as_sitesearch=www.t10.org> yields no clearly relevant hits.

> No one sees the need
> to extend this abomination to the 16 byte CDB group.

Could someone explain how "everyone knows" Seek is an abomination that can be effectively discouraged by carefully not mentioning that op x8B is not Seek(16)?

I've also missed why it was a GoodIdea to offer 16 byte Cdb's without offering any vendor-specific 16 byte Cdb's - is that a related confusion?

> The Seek(6) and Seek(10) are
> not used except for test cases.

The de facto standard - Wintel - uses x2B Seek(10) in the field.

Myself, I think this usage is brain-dead.  I've seen it on an Atapi bus that does not support disconnect when the device ID is not 0 - we can all hope it appears nowhere else.

Anyhow, there, a Seek precedes every x28 Read(10) or x2A Write(10), so that the host can slowly poll (19.2Hz) to see seek-complete, rather than leaving the bus busy during that time.

This approach adds an effective average adds about 0.5/19.2 = 26ms to every seek ... unless you work hard to fix it.  I think I remember these Seek's happen only if the ID of another device indeed is 0, but then they happen no matter whether that other device is in use or not.

> The Seek(6) and Seek(10) are
> not used except for test cases
> ... abomination ...

I have heard that we should count software that times Seek commands among the more pernicious of the lies, damned lies, and benchmarks with which ignorant but otherwise powerful people assault HDD vendors.

Perhaps this is the abomination meant here?

I've heard in response some HDD vendors treat Seek ops as no-op's.  I have myself seen Ata HDD that complete the first seek command immediately - you have to send another seek if you want to wait for the first to complete.

> not used except for test cases.

Maybe someone here can explain why In The Beginning Scsi bothered to include test tools?

I find the printed definitions disturbingly vague - I wonder how effectively they can be applied across more than one device vendor?  I have myself seen wildly varying interpretations of the x08 Caching and x01 ReadWriteErrorRecovery mode pages ... even creative pageLength's.

How can it make sense to obsolete Seek without also obsoleting that other stuff designed for diagnostic software?

> to obsolete Seek without also 

That query reminds me ... it seems we also are quietly not carrying forward x3E/3F ReadLong(10)/WriteLong(10) to become x9E/9F ReadLong(16)/WriteLong(16).

Thanks again in advance.    Pat LaVarre


*
* 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 mailing list