Best newfs under FreeBSD?

iang at iang at
Fri Dec 24 08:32:39 UTC 1999


Content-Type: text/plain; charset=us-ascii

"Robert J. Adams" wrote:
> We are building a new INN cnfs box using 50gig barracudas.. I was
> wondering what the best newfs options would be since this is going to
> have one large file per drive, I assume I don't need many inodes? I
> was thinking something along the lines of:
> newfs -b 65536 -m 0 -f 32768 /dev/blah
> Does the above look ok? Will it have any performance issues since it
> will be "optimized" for space and not time?

- - From the man page of tunefs:
     -m minfree
             Specify the percentage of space held back from normal
             users; the minimum free space threshold.  The default value
             used is 8%.  This value can be set to zero, however up
             to a factor of three in throughput will be lost over the
             performance obtained at a 10% threshold. Settings of 5% and
             less force space optimization to always be used which will
             greatly increase the overhead for file writes.  Note that
             if the value is raised above the current usage level, users
             will be unable to allocate files until enough files have
             been deleted to get under the higher threshold.

That said, I think that setting minfree to 0 will only affect file
creations and appending to existing files. If you dd the entire
buffer into the file system, I don't think that you'll have any
performance problems.

Changing the value for the frag size will also have little (probably
no) affect on performance since you're only going to have one file
in the file system. You could probably get away with leaving it as
default or as block size.

I don't think that changing the block/frag allocation size in the
file system is going to change performance much except maybe for
creating largeish files. If you're interested in getting improved
transfer rates from your file system I would suggest creating a
stripe and fiddling the interleave factor. I have found empiricaly
that ccd stripes are faster than vinum stripes. 

With a 5 way stripe with Seagate Cheetah LVD disks I get 54Mb/s
sustained writing a 1Gb file and 165Mb/s sustained reading a 1Gb

I would suggest doing some experiments with different block sizes
and stripe sizes for the file system and dd, bonnie or iozone (which
are in the ports collection).

- - -- 
Ian Freislich
UUNET S.A. (Pty) Ltd

Version: PGPfreeware 5.0i for non-commercial use
Charset: noconv


More information about the inn-workers mailing list