On Wed, May 10, 2000 at 08:25:08AM -0400, Dave Sill wrote:
[snip]
>
> >This is the problem, yes. You seem to have one inode per 4kbyte of
> >diskspace. This should always be sufficient.
>
> It's obviously not in this case. His 100,000 messages are apparently
> pretty small. But regardless of their size, you need 100,000 inodes
> to store 100,000 files. Each message in the queue requires at least
> two inodes.
I looked at one remote message and it took up 3 inodes: mess, info and
remote.
> >Hmm this is problematic. I just realized that for a disk to run out of
> >space before it runs out of inodes with qmail you need 1 inode per 1k.
>
> How do you figure that? The ratio of inode space to data space (for
> most filesystems) is determined at the time the filesystem is
> created. I don't see anything magic about 1k/inode.
The 'mess' would normally be a couple of kbytes, but info and
remote are under 50 bytes, normally. Averaging this I come to 'couple of
kbytes' (let's take 4), adding 2x50 bytes to this, gives me 4096+50+50=4196
bytes, divided over 3 inodes. 4196/3=1398.66. To have enough inodes for
this you need 1 inode per kbyte.
I have already been thinking in what formulation this would fit in LWQ - I
figured this is _the_ chance for me to contribute :)
Greetz, Peter.
--
Powered by WUT? - Peter van Dijk [student:sysadmin:developer:madly in love]
| `Yes, this was actually a hack and not | (petervd@|www.)vuurwerk.nl
| a scritp kiddie clicking a mouse button.' | www.dataloss.net
| - hackernews.com, commenting on the apache.org deface
