On Fri, 12 Feb 1999, Matti Aarnio wrote:
> Billy Harvey <[EMAIL PROTECTED]> wonders:
> > Benno Senoner writes:
> ...
> > > but actually the customer , in the worst case (in case of an fsck) must
> > > turn on the power of the machine 30-60min (for a 90% full 70GB array)
> > > before actual usage.
> > > In my case the machines can not remain powered up all the time, because
> > > the machines will be used for presentations at different locations.
> >
> > I'm new to raid discussion. Why would you expect a 60 minute fsck
> > everytime? Would the boot up not skip that if the shutdown was clean?
> > Journaling seems like a complicated solution to save the time of an
> > occassional fsck. Am I missing the obvious here?
>
> Not every time, but if the user has made some mistakes at
> the system shutdown time (it has been unclean), only
> journalled systems (to a certain degree) allow fast
> recovery.
[... this and other comments deleted ...]
You all miss the obvious point here, IMHO. The point of a journalled fs is
not that you save some time for fsck. Actually, even with a journalled
fs, I'd suggest doing an fsck from time to time just to be sure everything
is
fine.
The problem with a non journalled fs is that if you machine crashes or is
not properly shutdown, the fs becomes inconsistent, that is, you might
loose files. These might be files just created 1 minute b4 the crash
(shouldn't be a problem) but also be a file you had already worked on for
one week cpu time, it was just being modified during the crash and so it
is lost, or a full directory of files may be lost because someone added a
new file to this dir during the crash.
The fsck does not only need time, but fsck can not fully reconstruct the
fs. That is, fsck might have to remove broken info to get at least a
consistent fs w/o a chance to reconstruct all data. Well, from experiemce,
ext2 appears rather robust to me, I did not loose massive data due to
fsck's of unclean fs's yet. But there is no guarantee. IN theory, after a
crash during a critical fs operation, fsck might return with an empty fs.
(Well, at least it is a clean and consistent fs then ;-) ).
A journalled fs guarantees, that you can reconstruct the consistent state
of the fs at something like 1 minute or so before the crash. THIS is the
point about a journalled fs. It guarantees your data is not lost during a
crash, or better, that only the changes of the last minutes are lost, but
nothing more.
The need not to run fsck after a crash is only a nice side effect.
Michael.
--
Michael Weller: [EMAIL PROTECTED], [EMAIL PROTECTED],
or even [EMAIL PROTECTED] If you encounter an eowmob account on
any machine in the net, it's very likely it's me.
