[lkml]   [2001]   [Nov]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: Journaling pointless with today's hard disks?
On Sat, 24 Nov 2001 15:29:05 -0800, you wrote:

>Stephen Satchell wrote:
>> It is the responsibility of the power monitor to detect a power-fail
>> event and tell the drive(s) that a power-fail event is occurring. If
>> power goes out of specification before the drive completes a commanded
>> write, what do you expect the poor drive to do? ANY glitch in the write
>> current will corrupt the current block no matter what -- the final CRC
>> isn't recorded. Most drives do have a panic-stop mode when they detect
>> voltage going out of range so as to minimize the damage caused by an
>> out-of-specification power-down event, and more importantly use the
>> energy in the spinning platter to get the heads moved to a safe place
>> before the drive completely spins down. The panic-stop mode is EXACTLY
>> like a Linux OOPS -- it's a catastrophic event that SHOULD NOT OCCUR.
Correct, sort-of. The storage is not allowed to corrupt any data that
is unrelated to the currently active operation, (ie adjacent tracks or
sectors). Of course write-caching is asking for trouble.
>There is no "power monitor" in a PC system (at least not that is visible
>to the drive) -- if the drive needs it, it has to provide it itself.
>It's definitely the responsibility of the drive to recover gracefully
>from such an event, which means that it writes anything that it has
>committed to the host to write;
Correct. If a write gets interrupted in the middle of it's operation,
it has not yet returned any completion status, (unless you've enabled
write-caching, in which case, you're already asking for trouble) A
subsequent read of this half-written sector can return uncorrectable
status though, which would be unfortunate if this sector was your
allocation table, and the write was a read-modify-write.

>anything it hasn't gotten committed to
>write (but has received) can be written or not written, but must not
>cause a failure of the drive.
Reading a sector that was a partial-write because of a power-loss, and
returning UNCORRECTABLE status, is not a failure of the drive.

>A drive is a PERSISTENT storage device, and as such has responsibilities
>the other devices don't.
>Anything else is brainless rationalization.
> -hpa

Do You Yahoo!?
Get your free address at

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:13    [W:0.107 / U:0.432 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site