sbp, camcontrol, and Tagged Queuing

[Bob Johnson]

On Thursday 17 March 2005 10:08 pm, Drew Tomlinson wrote:
> I posted this a while back and am still having the same problem.  Can
> anyone offer any insight as to if the sbp man page suggestion about tagged
> queuing is something I should try?  Is there any risk of screwing up my
> drives by trying this?
>

Tagged queueing queues up multiple instructions for the drive simultaneously.  
The drive then attempts to sort them out and execute them in optimum order. 
Some drives that claim to support tagged queueing do not correctly do so, and 
don't perform well when it is used (and may lose data).  If you set the queue 
size to one, as recommended in the passage you reference, then only one 
instruction will be issued to the drive at time, and it will behave like a 
drive without tagged queueing.  It will do no harm to the drive.  If the 
drive correctly implements tagged queueing, this will slow down the drive, 
but if it does not correctly implement it, then this may dramatically speed 
up the drive (and make it more stable).  I have an external drive that 
manages 1.3 MBps transfers with queueing enabled, and 25 MBps transfers when 
I set the queue size to one.

As for whether it will help your specific problem, I don't know, but I can't 
see how it would do any harm to test it.

This issue is not specific to FreeBSD.  Any OS that supports tagged queuing 
has problems with some drives.

- Bob


> Thanks,
>
> Drew
>

[...]

> da2 and da3 are two IDE drives in a firewire enclosure.  These are also
> the drives that come up "referenced" after restarting.  What do these
> errors mean?  How can I correct them?  Is the following section from the
> sbp man page applicable to my situation?
>
> Some (broken) HDDs don't work well with tagged queuing. If you have prob-
> lems with such drives, try ``camcontrol [device id] tags -N 1'' to dis-
> able tagged queuing.
>
> Thanks for your help!
>
> Drew
>