On Mon, Sep 03, 2001 at 01:15:49PM -0400, Thor Lancelot Simon wrote:

 > We should actually benchmark filesystems with many *less* cylinder groups
 > (32K filesystems) against those with about 50 cylinder groups on modern
 > disks to see which way we handle real workloads better.  However, I think
 > there's plenty of evidence to support switching to 16K blocks and as many
 > cpg as we can get, given the geometry (about 300 for most new disks) right
 > now.

I agree with pretty much everything you have said (not a surprise, since
you and I have been talking about this stuff off and on for ... years now :-)

However, I think we should put some of the smarts into newfs_ffs.  For
things like flash devices (which show up as disks), we want small numbers
of cylinder groups mostly because it's needless overhead.  For large, fast
disks, we want small numbers of cylinder groups.  But for those of us that
still have small, slow disks connected to our Sun3s, VAXes, and DECstations,
we'd like to have something that like the current policy.

Note, for flash device, you also want smaller block sizes (because the
devices tend to be "small", e.g. 2-8MB).

So, for the larger the file system you're creating, newfs_ffs bumps the
blocksize and the cpg using some hueristic (coming up with the correct
hueristic is left as an exercise for the reader).

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>