On Thu, 19 Jan 2006, PFC wrote:

	This isn't really a md issue, but it's really annoying only when using
RAID, because it makes a normal process (kicking a dead drive out) so slow
it's almost non-functional. Is there a way to modify the timeout in question ?

yeah i posted to l-k about similar problems a while back... i've got a 
disk which boots fine but fails all writes... useful for showing just how 
bad the system can become with a dead/dying disk.

western digital is selling "raid edition" disks now -- and part of their 
marketing material discusses the long timeout which commodity disks 
implement <http://www.westerndigital.com/en/library/sata/2579-001098.pdf>.  
the raid edition disks give up earlier on the assumption the raid layer is 
going to take care of things.  it's really too bad this isn't just a 
tunable parameter of the disk.

even still -- the linux kernel could probably do something about this... 
drivers could have a blockdev(8) tunable timeout, and a mode where the 
driver just gives up entirely on the device at the first error/timeout and 
return EIO for all outstanding requests at that point... and the driver 
could remain in this state until an explicit request to re-attempt normal 
operations.

-dean