Possible to preallocate files that always begin/end on stripe unit/width?

[Green Guy]

I am working on an app that will write to preallocated files. I can control
how big the files are (as long as they are between 4-6GB) and the amount of
data sent with each write() call.
I have a 3.2TB virtual drive that I need to use as much of as possible but
performance is the number one concern.

The system is an 8 disk RAID5 with a 256k stripe. Based on this, I am using:

mkfs.xfs -b size=4096 -l version=2,sunit=512 -d su=256k,sw=7 -f /dev/sdb
meta-data=/dev/sdb               isize=256    agcount=32, agsize=26679232
blks
         =                       sectsz=512   attr=0
data     =                       bsize=4096   blocks=853735424, imaxpct=25
         =                       sunit=64     swidth=
 blks, unwritten=1
naming   =version 2              bsize=4096
log      =internal log           bsize=4096   blocks=32768, version=2
         =                       sectsz=512   sunit=64 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0

and
mount -t xfs -o sunit=512,swidth=3584,inode64,nobarrier,logbufs=8  /dev/sdb
/mnt

Based on what I have read, it appears then my "most optimal" write/read size
would be 1835008 bytes (3584 *512).

So based on that, I want to preallocate files somewhere between 4-6GB each
that will always begin and end on the stripe unit and stripe width and be a
multiple of 1835008.

I know that XFS metadata such as the log, AG info, etc also need to be
accounted for, but I am not sure the best way to determine how much space
they will take.

I assume that the fact that, once I set the number of files on the disk,
they will not change/expand must be an advantage, but I am unsure how to
leverage it.

I have tried file sizes in several different multiples of 1835008, but for
all of them, bmap -vp has flags 00011.

How can I determine the correct file size and preallocate them in a way
where they always begin/end on stripe unit/width taking the filesystem
overhead into account?
Note: the log needs to stay on the same fs.

Thanks
g3k
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://oss.sgi.com/pipermail/xfs/attachments/20100903/54ec301e/attachment.htm>