logical volumes under 3.3

[David A Higgen]

In article <13693 at hydra.gatech.EDU>, robert at shangri-la.gatech.edu (Robert Viduya) writes:

> Has anyone who's gotten 3.3 played with this stuff yet?  Is there a limit
> on how large logical volumes can be?

Guess this is my baby, since I'm the logical volume originator.
Yes, logical volumes (and filesystems thereon) may exceed 2 gig, though 
the size of regular files is still limited to 2 gig by the 'signed long'
nature of the lseek argument. (And the fact that regular filesize is stored
internally as a signed long, which precludes playing any tricks with
SEEK_CUR or SEEK_END).

There is actually an 8 gig limit to the size of a filesystem, imposed by
some data structures in the on-disk filesystem stuff. Logical volumes
themselves have no (practical) limit, but 8 gig is the maximum size of
filesystem you can put on one. Oops, I don't think this is documented
in the mkfs man page... mea culpa for that: however, you will get a 
reasonably self-explanatory error message from mkfs if you try to make
a filesystem bigger than 8 gig. Or if you try to grow an existing one 
to over 8 gig with growfs (just a little plug here, folks: existing
filesystems can be grown without losing their contents in 3.3. Check out
growfs(1M)!).

Two further hints: if you're concerned with balancing disk usage, it's probably
worth setting your logical volume up as a striped volume; this will
distribute traffic evenly amoung your disks and give you a better overall
throughput in a multiuser situation. 
Also, with SCSI disks, you will get improved performance if you use the 
mount option to increase the filesystem "logical block size" to larger
than its default. See the mount(1m) man page: basically you should add
"lbsize=65536" to the mount options. 



			Dave Higgen    (daveh at xtenk.asd.sgi.com)