On Feb 14, 2011, at 3:17 PM, David Brodbeck wrote:
> I would be curious to hear stories from people who actually *have* run
> into SSD failures related to write limitations.  I've heard a lot of
> speculation but no actual anecdotes.  I'm sure they're out there; but
> I also know people are more likely to complain when things go wrong
> than talk about things going right, so my suspicion is it must be
> rare.

Back around 2005 / 2006, we were using a bunch of Soekris 4511's, IIRC, running 
NetBSD and a network IDS we'd been working on, which possibly generated 100s of 
MB to a few GB of logging per day.  Whoever did the initial setup didn't 
realize that the flash cards of that timeframe were limited to 10K writes or 
so, and after a few months you started getting 16K chunks of old logfile data, 
or 16K chunks of new and old logfile data corrupted together-- looked to be a 
binary OR of the 0 bits.

Nothing reported that writes were failing-- evidently the flash cards didn't 
notice an error and thus didn't report it back to the system.  Switching /var 
to tmpfs resolved the issue for us.

>From what I understand (a quick review of wikipedia helps :), modern flash 
>cards are now typically rated for 100K writes, include ECC bits to actually 
>correct or at least detect errors and try to remap bad blocks to unused 
>blocks, and implement wear-leveling techniques of varying degrees of 


PS: Reposted from a NetBSD thread, was 
freebsd-questions@freebsd.org mailing list
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"

Reply via email to