26 Sep
2008
26 Sep
'08
5:58 a.m.
Chuck Guzis wrote:
So, maybe I should put one of the "industrial
strength" CFs to the
test on a mailserver. For purposes of testing, I could hook on a
hard disk and back new mail up to it periodically. If the CF bites
the dust suddenly, at least I wouldn't miss much.
Surely this is something where you can simulate things in a condensed time
period by scripting repeated file read/write/creates of random pathnames and
sizes? i.e. the simulation can match the real-world conditions pretty well,
but hopefully show up problems sooner rather than you having to potentially
wait for months.
The only gotcha I can see is that so many contiguous ops might cause
heat-related failure that might not be seen in the wild, but that's probably
no big deal.
As I understand things, the "industrial" CFs
use SLC and implement
real wear-leveling algorithms (although no manufacturer will disclose
what said algorithms might be).
I think they normally quote a mean 'write ops before failure' figure, don't
they? That might be useful in a comparison against rotating storage, if you
can pull some figures for how often you write to your current storage system.
I started running from a CF card on my main data recovery box back in the UK,
but never got to the point of hammering it enough to see any failure before I
moved. That was booting from a lowly 32MB card though, so CF lifetime wasn't
an issue - I just kept a raw 32MB image of the card on my fileserver and it
would have been trivial to toss a new card into the system and blat the image
across if it died on me (the card only held the OS and utils, so I essentially
ran it read-only)
cheers
Jules