I have a server I just set up with an Adaptec 2400 IDE RAID controller,
and I'm a bit curious about the filesystem parameters I should use
to set it up.  The default parameters from dmesg are 63 sectors/track,
255 tracks/cylinder.  This results in some interesting error messages
from newfs:

    # newfs /dev/rld1a
    Warning: insufficient space in super block for
     rotational layout tables with nsect 63 and ntrak 255.
    File system performance may be impaired.

This newfs doesn't work at all unless I boost the block size to 32KB
and the fragment size to 4KB (from the default of 8KB and 1KB).
How much should I worry about this message?  Using 63 sectors/track 
and 64 tracks/cylinder eliminates the message, but what else
does it do to the overall performance?

I tried a couple of alternate configurations:

Sectors/Tracks/Cylinders	Block/Frag	tar	copy
63/255/9729			32K/4K		70.69	4.63
63/64/38765			32K/4K		73.85	4.54
63/64/38765			8K/1K		59.32	11.24

The tar test was:

     # /usr/bin/time tar xzf /another-disk-on-the-2400/gnusrc.tgz

The copy test was similar, but copied over about 10 files of NetBSD
install stuff, totalling around 194MB.  (Most of the source data
should come from cache, as the system has 1GB of RAM.)

This totally confounded my assumptions -- I figured that using the
larger disk blocks would be a win for almost all writes to the disk.
I guess untarring the file resulted in lots of smaller files that
required seeks past 'em, or allocating the fragments was especially
expensive.

I'm planning to store lots of files of three to 10 MB apiece on 
this filesystem, so I suppose I'll stick with the larger blocksize for those.

However, I have another disk on this controller that I plan to use for
CVS.  Perhaps I should use the smaller blocksize for this one?
Did my simplistic tests miss some important considerations?  I'd love
to get a pointer or two to instructions for tuning FFS on large disks!

Thanks!

Kevin
kml@selresearch.net