[Dirvish] Paul's JFS experiment ( was: dirvish write error )
dhoworth at mrc-lmb.cam.ac.uk
Thu Jul 28 02:24:26 PDT 2005
Keith Lofstrom wrote:
> On Wed, Jul 27, 2005 at 08:57:50PM +0200, Paul Slootman wrote:
>>I've been doing some tests on a box with 24 400GB SATA disks... ext3
>>doesn't even create filesystems larger than 8TB :-) reiserfs 3 oopses
>>the kernel, reiser4 works fine, but I'm probably going with jfs as that
>>gave the best performance.
> Let us know how that works out. JFS is better than ext3, because
> it has dynamic inode allocation (you don't run out of inodes). I
> suspect it is safer and faster than reiserfs, but it does have
> fixed block sizes, and I suspect that is less optimal for
> directory-rich, heavily-hard-linked dirvish vaults.
> I wonder if there is some kind of tool that can traverse a filesystem
> and estimate what percentage of the space is occupied by directories?
> For those of you wondering what we are talking about, here is an
> overview of JFS, IBM's Journaling File System :
FWIW, SuSE withdrew support for JFS in SuSE 9.3, citing technical
problems. I don't know any more than that.
More information about the Dirvish