1 |
Am 08.01.2013 20:53, schrieb Grant Edwards: |
2 |
> On 2013-01-08, Pandu Poluan <pandu@××××××.info> wrote: |
3 |
>> On Jan 8, 2013 11:20 PM, "Florian Philipp" <lists@×××××××××××.net> wrote: |
4 |
>>> |
5 |
>> |
6 |
>> -- snip -- |
7 |
>> |
8 |
>>> |
9 |
>>> Hmm, good idea, albeit similar to the `md5sum -c`. Either tool leaves |
10 |
>>> you with the problem of distinguishing between legitimate changes (i.e. |
11 |
>>> a user wrote to the file) and decay. |
12 |
>>> |
13 |
>>> When you have completely static content, md5sum, rsync and friends are |
14 |
>>> sufficient. But if you have content that changes from time to time, the |
15 |
>>> number of false-positives would be too high. In this case, I think you |
16 |
>>> could easily distinguish by comparing both file content and time stamps. |
17 |
>>> |
18 |
>>> Now, that of course introduces the problem that decay could occur in the |
19 |
>>> same time frame as a legitimate change, thus masking the decay. To |
20 |
>>> reduce this risk, you have to reduce the checking interval. |
21 |
>>> |
22 |
>>> Regards, |
23 |
>>> Florian Philipp |
24 |
>> |
25 |
>> IMO, we're all barking up the wrong tree here... |
26 |
>> |
27 |
>> Before a file's content can change without user involvement, bit rot must |
28 |
>> first get through the checksum (CRC?) of the hard disk itself. There will |
29 |
>> be no 'gradual degradation of data', just 'catastrophic data loss'. |
30 |
> |
31 |
> When a hard drive starts to fail, you don't unknowingly get back |
32 |
> "rotten" data with some bits flipped. You get either a "seek error" |
33 |
> or "read error", and no data at all. IIRC, the same is true for |
34 |
> attempts to read a failing CD. |
35 |
> |
36 |
> However, if you've got failing RAM that doesn't have hardware ECC, |
37 |
> that often appears as corrupted data in files. If a bit gets |
38 |
> erroneously flippped in a RAM page that's being used to cache file |
39 |
> data, and that page is marked as dirty, then the erroneous bits will |
40 |
> get written back to disk just like the rest of them. |
41 |
> |
42 |
|
43 |
Related: The guys in [1] observed md5sums of data and noticed all kinds |
44 |
of issues: bit rot, temporary controller issues and so on. |
45 |
|
46 |
[1] Schwarz et.al: Disk Failure Investigations at the Internet Archive |
47 |
http://www.hpl.hp.com/personal/Mary_Baker/publications/wip.pdf |
48 |
|
49 |
Regards, |
50 |
Florian Philipp |