overhead with jfs filesystem

[Todd Bowden]

I have a problem I know has a workaround.

I have 2 filesystems that I created both are were made using the same options except largefiles. Here is an ouput of a working machine.

mkfs -F vxfs -m /dev/sapvg14/sapdata10
mkfs -F vxfs -o ninode=unlimited,bsize=8192,version=3,inosize=256,logsize=256,la
rgefiles /dev/sapvg14/sapdata10 8704000

/ # bdf /oracle/PRD/sapdata10
Filesystem kbytes used avail %used Mounted on
/dev/sapvg14/sapdata10
8704000 6269440 2415544 72% /oracle/PRD/sapdata10

Here is the other machine:

/oracle/PRD/sapdata10 # mkfs -F vxfs -m /dev/sapvg14/sapdata10
mkfs -F vxfs -o ninode=unlimited,bsize=8192,version=3,inosize=256,logsize=256,nolargefiles /dev/sapvg14/sapdata10 8704000

/oracle/PRD/sapdata10 # bdf .
Filesystem kbytes used avail %used Mounted on
/dev/sapvg14/sapdata10
8704000 2552 8633472 0% /oracle/PRD/sapdata10

The difference is 48960k. Why does this show like this and can I change it? How?

Thanks for any help you can provide

Todd

[Rainer_1]

your difference results in the usage of the filesystems.
The first fs has 72% usage and the second 0%
For every directory or file the fs needs to create an inode entry.
This increases the fs usage.
You cannot change this behavior.

[Todd Bowden]

Maybe that filesystem was a bad example. I have other filesystems that are at 100% as:

/ # bdf /oracle/PRD/sapdata8
Filesystem kbytes used avail %used Mounted on
/dev/sapvg06/sapdata8
7987200 7984408 2792 100% /oracle/PRD/sapdata8

/ # mkfs -F vxfs -m /dev/sapvg06/sapdata8
mkfs -F vxfs -o ninode=unlimited,bsize=8192,version=3,inosize=256,logsize=256,la
rgefiles /dev/sapvg06/sapdata8 7987200



If you add the used and available they add up to kbytes, so why doesnt it add up on the other filesystems?

I know I did something to get it so a user can use all available disk space. Im just not sure what it was.

[John Palmer]

Hi,

There was a discussion about this some time ago, have a look at:-
http://my1.itrc.hp.com/cm/QuestionAnswer/1,1150,0xec707e990647d4118fee0090279cd0f9,00.html

Regards,
John

[Bill McNAMARA_1]

If you create a filesystem with inodes =
filesystem/8k you should get the max size
possible reported from df.

With JFS there is no real concept of block size/inode, rather physical extent/inode.
JFS also dynamically creates inodes on demand,
when fsystem is getting tight, you'll end
up with a lot of defrag and non contiguous
file growth. Can't reall be avoided without
defrag/Online JFS