On Tue, Mar 12, 2002 at 02:34:27PM +0100, Wojciech Puchar wrote:
> > Unused inodes shouldn't take any time to fsck!
> 
> reading them too?
> yes it takes longer to scan 10 milions of inodes than 1 milion.
> 

Lets see, disk inode is 128 bytes, so a 160Gb drive would have
160Gb/4k = 40M inodes, taking 5Gb of disk, at 20Mbyte/s this
would take 250secs.

I would have thought that the directory reads and extension
blocks would take longer - since they are not contiguous
reads.

However I ought to be possible to avoid reading inodes that
have never been allocated - does depend on the allocation policy.

I wonder how clever fsck is?  After all it has a significant
amount of memory available on many systems.  It ought to be 
able to avoid most of the seeks?

> > FWIW I changed the parameters when creating my root fs, and it runs
> > out of inodes!  I can't remember the paramters I used, but I changed
> > the cylinder group size and bytes/inode.  I do wonder if the sum
> > allowed for both since I don't recall requesting 32k/inode.
> >
> if you have small root fs you shouldn't use such option.

Define small!  I reconned root would mainly have program binaries
in it.  So 2k/inode seems a waste of space.  Indeed my other
root fs is 70% full using 4% of its inodes.
Indeed, it is only because the new root fs is 2Gb that it had
the space for me to dump a build object tree into it - that
does have rather a lot of small files :-(

	David

-- 
David Laight: david@l8s.co.uk