On Fri, Apr 01, 2011 at 06:54:37AM +0200, Otto Moerbeek wrote:
> > A request to tech@,
> > Can you please consider architecture wise bumps in filesystem defaults
> > to data block size and fragment size such that inodes to be scanned
> > are reduced? When you run fsck you really want it to be correct and if
> > possible faster? I don't know when this fsck diff will get integrated
> > (it is early in cycle!!! wink wink) but defaults can get bumped after
> > discussion.
> >
> > Thanks in advance
>
> I think the defaults are reasonable.
Some data points to stress that.
Filesystem Size Used Avail Capacity iused ifree %iused Mounted on
/dev/sd0i 1.3G 772M 479M 62% 88579 93307 49% /usr/src
/dev/sd1e 133G 2.6G 124G 2% 265879 8559975 3% /scratch
A small (16k block) filedssytem and a largish (32k block) filesystem.
All created with defaults. The /scratch filesystem has a few src tree copies
on it.
You see that if I would create less inodes, in both cases I would run
the risk of running out of inodes before running out of space. That
would suck.
In your particular case it might work, but the defaults assume that
running out of inodes is a really bad thing you want to avoid.
Also, moving to larger blocks wastes more space. So that also a reason
to be conservative.
-Otto
