I agree; OCFS2 is unappreciated, and it is quite a good filesystem. However, it does have it's limitations.
We use OCFS2 for backup volumes on our fibre channel SAN storage. We've found it performs adequately enough, although filesystem performance can slow down when you have a lot of sequential writes going to the storage.
It does some things really well, but in the end it has a single point of failure - the storage array or server hosting the storage becomes a bottleneck and SPOF. It doesn't take the place of newer clustering technologies that spread data around multiple points of failure.
OCFS2, in the v1.4 branch has some extreme fragmentation problems. It seems to be just fine if you're dealing with large files, or lots of small files that don't change frequently. However, if you've got lots of turnover of lots of small files, you're going to get bitten.
If that describes you, you need to be willing to track the latest kernel releases, and able to rebuild your FS semi-frequently.
(I'm being bitten by this right now, and I need to rebuild the FS... not something I'm happy about.)
We ran into this as well. The main problem is that when you get fragmented too badly you run into space allocation issues where you have tons of free space, but the kernel returns an "out of disk space" error because it couldn't allocate new blocks.
The only solution is to rebuild the filesystem and restore from tape. I suppose Oracle doesn't really have any motivation to fix this because most of their customers are using it for large database files rather than small files.
It's also annoying to have open bugs that can't get any traction... I'm more than willing to do almost any test (short of destroying my data) to help them along, because an easy way to get multiple boxes with block access to a volume is pretty awesome...
We use OCFS2 for backup volumes on our fibre channel SAN storage. We've found it performs adequately enough, although filesystem performance can slow down when you have a lot of sequential writes going to the storage.
It does some things really well, but in the end it has a single point of failure - the storage array or server hosting the storage becomes a bottleneck and SPOF. It doesn't take the place of newer clustering technologies that spread data around multiple points of failure.