On Tue, Jan 18, 2000 at 06:07:28AM +0100, Andreas Dilger wrote:
> [note email sent to linux-fsdevel, as it is OT for l-k]
> Jamie Lokier writes:
> > Alan Cox wrote:
> > > Its caught errors before that probably avoided people losing data. Its a
> > > good sanity check. The "stupid" defaults are two other things
> > >
> > > o Using 1K blocks on large disks (4K is way faster) and
> > > 4K checks way faster too
> > >
> > > o Not using the "sparse superblock" option on large disks
> > > when creating them.
> > >
> > > Unfortunately I dont think there are any "in place" fixers for those
> > > creation time choices.
> >
> > Sparse superblocks can be turn on/off by the current tune2fs.
> > You follow it by a fsck to clean up, then it's done.
> > I was hoping they'd speed up the mount but its still takes an age...
> >
> > Why does 4k blocksize check faster? (I've never tried it). Does it use
> > fewer superblocks?
>
> Not only are there 1/4 fewer (superblocks, group descriptors, inode/block
> bitmaps) with 4k ext2 filesystems, but also files have fewer blocks,
> including single-, double-, and triple-indirect blocks, so much less checking
> for a given amount of data.
And much more waste. I recently discovered that a ~5GB 4K ext2 source
code partition has ~480MB wasted in file tails. The same fileset
(about 192k files) would only waste ~108MB with 1K blocks.
So sometimes 1K blocks or file tail packing (like in UFS or reiserfs)
are a good idea. Unfortunately both are somewhat discouraged in 2.3 (the
new IO paths seem to be heavily optimized for page sized IO only)
-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Jan 23 2000 - 21:00:17 EST