You (Svein Skogen) wrote:
> > Check:
> >
> > http://storagemojo.com/2007/09/19/cerns-data-corruption-research/
>
> Which is another worthless document, without any info on what
> controllers they actually tested with.
>
> I too can make claim about corruption and data loss. But without
> pointing the finger to what was actually tested, that's just a claim,
> nothing more.
>
> //Svein
What's it, your after? That study is more than 2 years old now, and the
RAID-controllers, that didn't work, probably will have updated their FWs by
now.
And: The Authors are listed, so, go ask them, if you really are interested in
those details.
More importantly is the fact, that bit-errors DO happen, and as I stated
earlier, with the average failure rate and storage size of today it's close to
100% certainty, that if you have more than 128 TB of data, you have an
error. That's a simple derivative from the infos, available from the
manufactures themselfes:
For example, for the Seagate ST31000640SS (1 TB SATA):
AFR: 0,73%
Nonrecoverable Read Errors per Bits Read: 1 sector per 10E15
Error Control/Correction (ECC): 10 bit
>From AFR: If you have 200 disks, 1.46 die per year.
ECC: If you have 1024 errors, one will NOT be detected
After that: 1 sector per 10E15 Bit ~= 128 TB is defect
More precisely: 10E15 = (10E3)E5 = 1024E5 Bit = 1024 Terabit = 116.4
Terabyte => There's a defect sector every 116.4 Terabit.
These are the DOCUMENTED ERROR RATES of standard consumer disks! Take
enterprise disks, and it's around 1 sector per 10E16 bit and an AFR of 0.55%
(Seagate Savvio 15K.2 ST9146752SS). So, not as bad as above, but still there
are documented possibilities of errors!
And, therefore, it's now a necessity to do error checking at every level!
Matthias
--
Matthias Pfützner | Tel.: +49 700 PFUETZNER | Klaus Kinkel war in China
Lichtenbergstr.73 | mailto:[email protected] | und wollte die Menschen-
D-64289 Darmstadt | AIM: pfuetz, ICQ: 300967487 | rechte sprechen, aber die
Germany | http://www.pfuetzner.de/matthias/ | waren nicht da.
_______________________________________________
opensolaris-discuss mailing list
[email protected]
