Re: Big filesystems.
Jeffrey W. Baker wrote:
And the only common thread to all of these is XFS. You can say that
maybe NFS is the problem, maybe this or that is the problem but the fact
is that all the other filesystems survive in the environment and not
XFS.
I never implied NFS was the problem. What I did state was saying that
XFS is unsafe over MD or LVM because NFS against a XFS filesystem causes
corruption doesn't follow.
Both of the problems I mentioned were XFS specific bugs. I never
implied they were NFS releated.
Note that CIFS causes the same corruption that NFS does. Is Samba also
broken?
No, I never implied that.
Note that XFS will corrupt a loop device on any kernel. I guess the
loop device isn't "below" XFS in your world view?
Proof?
XFS is broken because it uses its own buffer management layer, imported
to the tune of 1000s of lines of code from Irix. It maybe be in the
kernel but it doesn't use kernel facilities and/or it uses them
incorrectly.
The pagecache layer has more or less been fully integrated as of 2.6.
It was the reason it wasn't initially imported in 2.4, but lots of work
went on to make it a complete member of the Linux kernel.
If you want to have a very large filesystem with lots of files *and* you
are only going to use it in the "normal" fashion - directly attached
storage not exported via NFS or CIFS - *and* you are only going to use
RHEL or SuSE kernels (not kernel.org) then you will probably be fine.
But you should realize what you are getting into.
I run a debian kernel exporting data via CIFS every day to myself and
others, frequently pushing > 10Mb/s against several files with no data
corruption whatsoever.
Without more details, there's no way to know if you what you were hit
are *known* bugs and therefore expected behavior.
While unfortunate, they are a reality of using any filesystem but
ext2/ext3 on Linux (and and occasional reality of even them).
Adam
Reply to: