[ale] Need an iso to wipe hard drives.

[Greg Freemyer]

On Wed, Apr 21, 2010 at 12:16 AM, JK <jknapka at kneuro.net> wrote:
> On 4/17/2010 8:20 PM, Doug McNash wrote:
>>
>> ---- Chris Fowler<cfowler at outpostsentinel.com>  wrote:
>>> I also used to have a program that I wrote which would open a file for RW and fill it with
>>> random stuff.  If would seek back to the beginning and repeat a few times then close and
>>> finally delete.
>>
>> Unless this file is much bigger than the RAM used for file buffer caching (which on linux is
>> quite a large percentage) not much is actually being written to the magnetic media unless you do
>> a sync, fsync or unmount the drive.
>
>
> Except that there's a kernel thread that writes dirty buffers
> to disk periodically.  The semantics of the "sync" system call
> are, "Write all dirty buffers to disk and don't return until you're
> done". But in normal operation I'd be astonished if most dirty
> buffers weren't written out within a few seconds, even without an
> explicit sync.  Only if something was writing data faster than
> it could be sent to disk, would there be much danger of having
> a lot of dirty buffers in RAM for any length of time.
>
> Don't know about hardware disk caches, but that's what happens
> in the kernel. (Unless things have changed a whole lot since
> 2.4.)
>
> -- JK

A very reasonable expectation.

Unfortunately there have been a few different kernel bugs that keep it
from resembling reality in any way shape or form.

A couple years ago, xfs I believe it was had a bug where the kernel
never did an automated sync.  So you could have 12 hour old data
sitting only in kernel cache with an idle machine.   ie. the more the
machine was used the less stale the cache because the new data pushed
the old data out, but with an idle machine data just sat in the cache
and never went anywhere.

With the current 2.6.34-rc kernels there is a ext4 bug that sync is
taking 20 minutes or so to complete with a busy machine.  What's
really bad about that is during that 20 minutes new i/o is blocked, so
a simple vi command can take 20 minutes to open a file!

At any rate, depending on the kernel to sync out its write cache every
30 seconds or so is the opposite of paranoid.

Greg