Patrick R. Wade wrote:
> Well, all you've really done is trade fsck for journal restore, which
> is also time-consuming, but at least is more reliable than fsck.
As the original xFS(*) design documents at SGI put it:
recovery time proportional to filesystem activity level
at the time of crash, independent of filesystem size
... or words very similar to those.
I just now had a fscking power failure at home (literally fscking, not
fscking as an euphemism for that other word), and spent about 20
minutes getting all systems booted and connected.
> One of the more interesting crash-recovery schemes i've seen is in
> ErOS (http://www.eros-os.org), which periodically saves a known-good
> snapshot of the running system, so that if you have a crash, instead
> of rebooting from scratch, it just restores from the known-good state,
> so e.g. your shell sessions pick up where you left off.
Hmmm. If it were snapped often enough, and rebooted quickly enough,
your TCP connections would survive, too. Cool!
* Trivial fact of the day: xFS originally had a lowercase x. During
project planning (1992-93?), they bandied a lot of acronyms for the
filesystem around, all ending in "FS". But they couldn't decide on
one. Finally, they decided to use "x" as a placeholder for the
first letter, which they left TBD. They were adamant that x didn't
stand for anything. (-:
--
Bob Miller K<bob>
kbobsoft software consulting
http://kbobsoft.com [EMAIL PROTECTED]
