[wellylug] USB 3.0 via PCI-e, & SATA HDD enclosures/docks

Ewen McNeill wellylug at ewen.mcneill.gen.nz
Tue Jul 19 22:26:43 NZST 2011


On 2011-07-19 20:51 , Cliff Pratt wrote:
> Yeah good points, but a) I thought that drives/caches retain enough
> power to write the last of the data in the cache and

FWIW, I have very little faith in this occurring in a meaningful 
fashion: the buffering is too big, and the drive staying up to useful 
speed to reliably write once power is removed is too short.  (See also 
write behind caching on by default in most mainstream devices, eg: 
http://brad.livejournal.com/2116715.html)

My normal approach is to umount the disk, and then wait "a while" 
(typically a minute or two or however long it takes to walk from my 
workstation to the machine in question) before removing power from it. 
(Which is very "sync<NL> sync<NL> sync<NL> halt<NL>" like I know, see eg 
http://www.classiccmp.org/pipermail/cctalk/2001-April/169861.html.)  I 
have at least one (powered) USB enclosure which seems to need this sort 
of approach: without giving it a bit of time the final "file system 
clean" write doesn't seem to get flushed to the media (it gets checked 
on the next mount), so I'll usually give it a minute or two after 
unmounting, then pull the USB only (leaving it powered), and give it 
another minute or two before pulling the power.  (Flash media, 
currently, has less write buffering depth than "spinning rust".)

So as noted earlier I'll do what I can to ensure that all the bits I 
want are committed to the underlying media.  But I draw the line at 
spinning down the drive first (especially since in most server 
environments it's difficult to issue a command like that anyway: there 
are layers of indirection in the way).

> I was also under the impression that even older IDE drives didn't
> actually park the heads.

I guess that depends on what you mean by "older" IDE drives, and whether 
you include "unloaded the heads" in "parked the heads".  (See, eg, 
http://en.wikipedia.org/wiki/Hard_disk_drive#Landing_zones_and_load.2Funload_technology) 
  AFAIK the "landing zone" approach is still pretty common, they just 
got better at reducing the chances of stiction, etc.

With older hard drives (eg, full height 5.25" ones) you could definitely 
tell the heads were being parked as the power removed (compared with 
older still ones before them which needed a "park the heads" command 
run).  Modern near silent hard drives make it difficult to be sure, by 
sound, what they're doing.  And I suspect many recent drives probably 
idle their heads over the landing zone if they've not been used in a 
while anyway.

Ewen



More information about the wellylug mailing list