[Soekris] Smallest possible MFS /dev on OpenBSD

[Bill Maas]

On Tue, 2008-01-15 at 14:56 +0200, Lars Noodén wrote:
> I've got OpenBSD 4.2 on a net4801 and am mounting /dev in RAM.  I'd like
> to reduce the amount of memory used even further.  Currently only %4 is
> used:
> 	Filesystem     Size    Used   Avail Capacity  Mounted on
> 	mfs:21800      853K   34.0K    777K     4%    /dev
> 
> What should I be aware of in /etc/fstab to minimize the memory footprint
> of /dev?  Currently, I have:
> 	swap /dev mfs rw,-P/dev.base,-i1,-s2300,-c264,-m5,nosuid 0 0

Couldn't resist .. here's ho far I got after about 45 mins. of testing:

# mount_mfs -s 512 -b 32768 -f 32768 -P/var/dev -o nosuid
swap /mnt/test/

# df /mnt/test
Filesystem  512-blocks      Used     Avail Capacity  Mounted on
mfs:17962          192       128        64    67%    /mnt/test/

# df -h /mnt/test
Filesystem     Size    Used   Avail Capacity  Mounted on
mfs:17962     96.0K   64.0K   32.0K    67%    /mnt/test/

Note: '-i' option is ignored here - it doesn't make any noticeable
difference at all. Block and fragment size seem to be the issues here.
With these set to the defaults, I never got past 13%. You would only
need to change the default bpi if you run out of inodes. Maintaining
fragments appears to be pretty disk-intensive, I suspect that if
(blk_siz == frag_siz (or whatever)) then it's turned off altogether.

I couldn't reproduce Jean-Yves' 23.0K (mount_mfs -i 256 -s 592
-P /var/run/dev swap /dev):

# mount_mfs -i 256 -s 512 -P/var/dev -o nosuid swap /mnt/test

# df -h /mnt/test
mfs:29060      143K   15.0K    121K    11%    /mnt/test

I suspect that the '-s' value 592 was a typo, since it was rejected over
here:
mount_mfs: reduced number of fragments per cylinder group from 72 to 64
to enlarge last cylinder group
mount_mfs: inode table does not fit in cylinder group

I can't find any hard data - apart from claims and statements - about
dynamic growth (nor shrinkage!) of the MFS partition, but should any
process hijack /dev e.g. as it's temp dir, then 96K (or 23, if
possible), should be a significant improvement over the 853K in Lars'
original posting, even if it's only a max. value.


Bill

-- 
"What's a computer?" - MES