[vcf-midatlantic] several things

Michael Brutman michael at vcfed.org
Thu Jan 2 21:36:07 EST 2020


I used to write enterprise class hard drive firmware for HGST before they
got bought by Western Digital.  Our code was also shard with the consumer
SATA drives.

Hard drives have *very* robust error detection and correction.  They will
generally go to extraordinary lengths to recover and relocate data on
marginal sectors.  The ECC algorithms implemented in the hardware are very
robust.  Drives do need to be powered up once in a while to do background
scanning for weak sectors, so just writing data and putting the drive on a
shelf is not the best idea.

As with many other things, the best approach is defense in depth (layers of
defense).  Store several copies of the data, and not on the same brand of
media or hard drive model.  Verify the data using a hash that was created
at the time the data was written to detect corruption, and use another copy
(that hopefully isn't also corrupted) to mitigate the problem.  Filesystems
like ZFS are awesome.

All of the big cloud data storage providers have your data stored in
multiple places; two copies isn't considered safe enough.  And of course
they spread them out geographically to mitigate damage from external
factors.


Mike


More information about the vcf-midatlantic mailing list