Inode density - newfs(8)

[Stephen Marley <stephen_(_dot_)_marley_(_at_)_catwoman_(_dot_)_cl-is_(_dot_)_com>]

Hi,

I have been using rsnapshot <www.rsnapshot.org> (which makes extensive
use of hard links) to keep incremental backups of several servers on an
OpenBSD 3.6 host.  It was all going well until I ran out of inodes on my
/usr filesystem, and as a result, I have a couple of questions.

Firstly, newfs(8) says that the default number of bytes per inode is 4k;
however, my df -i output suggests that the density is more like 1 per
8k. newfs.c also has some code that looks to me as though it assigns the
value NFPI * DFL_FRAGSIZE to a density variable, where NFPI and
DFL_FRAGSIZE are defined as 4 and 2048 respectively, and the cvs log has
some commits by tedu that say that the density was bumped. So, is
newfs(8) now inaccurate, and was halving the number of inodes really
such a good idea? Not all of us use large filesystems for storing mp3s
:¬)

Regardless of that, I need to increase the inode density to around 1 per
6k (or 4k to be safe, i.e. what it used to be!), and I am wondering what
the best way to go about it is. Am I right in thinking that the
only/easiest way to change this is through newfs, (single user mode ->
dump to tape -> newfs -i 4096 -> restore) or is there another way?

Are there any other optimizations I should consider for a filesystem
with a high ratio of inodes to data? 

Any advice is appreciated.

Kind Regards,
Stephen Marley

-- 
Network Systems Engineer, Datacentre Operations
Campbell Lee Internet Solutions <http://cl-is.com>