xfs agcount

frozenQueue · 12-09-2010, 06:13 PM

I'm building a Linux server that is going to have a bunch of VMs running on it, using the XFS filesystem. The mkfs.xfs man page indicates that increasing the number of allocation groups "implies that more parallelism can be achieved when allocating blocks and inodes." But what does that mean in practical terms. What are the positive and negative results of increasing the number of allocation groups?

Also, the man page indicates the default agcount value is "scaled automatically" depending on file size. What formula is used to determine that?

sag47 · 12-09-2010, 08:01 PM

I downloaded the xfs source via cvs.

While I look for the algorithm you can too. That's the beauty.

I checked out both linux-2.6-xfs and xfs-cmds. I ran the following command and found the man page you were looking at.

Code:

find . type f -print0 2> /dev/null | xargs -0 grep -iH 'scaled automatically' | cut -d: -f1 | sort -u

./xfs-cmds/xfsprogs/man/man8/mkfs.xfs.8

It's a start. I'm looking for the codebase for mkfs.xfs now. I'll update this post as I find more information.