On Wed, May 24, 2000 at 10:11:05AM +0900, enami tsugutomo wrote:
> Manuel Bouyer <bouyer@antioche.lip6.fr> writes:
> 
> > On Mon, May 22, 2000 at 04:40:03PM +0900, enami tsugutomo wrote:
> > > Why 10s is acceptable but 60s isn't acceptable?  To provide the way to
> > 
> > 5*10 = 50s, 5*60 = 5mn :)
> 
> Where 5* come from?  SDRETRIES?

It's the number of disks I have in the disk enclosure, for which I've seen the
problem.
But now that I think of it this computation is false: a timeout is likely
to trigger a reset on the adapter, and thus cancel other pending I/O.

> 
> > > cancel active i/o before removing the device is the right thing, isn't
> > > it?
> > 
> > Sometime the device removes itself without warnings :)
> 
> Hmm.  But, since it may be difficult to determine an exact least
> necessary timeout value (especially if device can accept multiple io
> request), I'm affraid that shortening timeout value causes bad
> effects.

I think 10s is reasonable for a READ/WRITE on a SCSI disk.

> 
> > I didn't notice PR #3009, thanks. Then would a different timeoutm, based on
> > fixed ws removable media be OK ?
> 
> ... and, I'm not sure the long timeout is necessary because of
> T_REMOVAL, T_OPTICAL or yet other each device specific
> characteristics.

I think it's because the device has some kind of power management and spins
down after some period of inactivity. I don't know of any fixed SCSI disk
which does this.

--
Manuel Bouyer, LIP6, Universite Paris VI.           Manuel.Bouyer@lip6.fr
--