[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[leafnode-list] Re: Run out of inodes
Am 18.05.2009, 15:46 Uhr, schrieb Gary R. Schmidt <grschmidt@xxxxxxx>:
> Matthias Andree wrote:
>> I'd suggest ext3 for Linux, UFS2 for FreeBSD, logging ufs on Solaris.
> Using ZFS for Solaris would be better, no pre-allocation of inodes and,
> if you use RAIDZ or RAIDZ2, the data is more secure.
> Oh, and ZFS is in the BSD kernel now, so it would probably be better
> there, too.
I do not believe in early adoption of file systems, particularly not if
the file system isn't the default for the respective operating system. Is
ZFS the default Solaris FS nowadays? If not, logging ufs is
well-established and works. It takes years to shake out file system bugs,
even if the system per se has been proven on one OS. Look at the XFS
history in Linux... it worked well in Irix, but took years to stabilize
Besides that, static vs. dynamic inode count is not the sole metric by
which to judge a file system, and switching file systems just because of
this one features seems short-sighted to me.
Reiserfs for instance has been hyped about a lot (no fixed inode count,
b(whatever)trees, tail packing - which I'd disable for workable speed)
with SUSE running file servers with it, but in the LANs I used to
maintain, it failed miserably, from the early 3.5 hype, through 3.6, and
by the time reiserfs had sort of stabilized and the tools had become
workable, I'd long since switched to ext3 with dir_index because I had
been fed up with all the self-destruct "features" of the new file system
Oh, if you're in for some fun, try NFS on the newest non-ext* filesystems.
I've had interesting experiences with NFS combined with either XFS or
reiserfs (and haven't even tried JFS).
The number of ZFS issues appearing on the FreeBSD lists is also big enough
that I'd stay away from it on BSD for a couple of years, and having tried
GEOM_JOURNAL + UFS2 - softupdates + async in FreeBSD 7.1 i386, I can only
say 'needs to mature before I'll touch it again'. I've run fsck -f more
than once until I got annoyed, rid myself of GEOM_JOURNAL and switched
back to UFS2 + softupdates - async.
Having purported all these non-scientific anecdotes, on to the consequence:
If someone wants to run leafnode on other than the default
semi-conservative file systems, feel free, but you should be able to tell
filesystem bugs from application bugs, i. e. you should know what you're
doing. I don't have time to support situations where file system bugs
punch their blunt trunks through leafnode's stomach.
leafnode-list mailing list