Re: Big filesystems.
On Tue, 2005-02-15 at 16:50 -0500, David Wood wrote:
> On Tue, 15 Feb 2005, Jeffrey W. Baker wrote:
> > That said, XFS is still your best choice if you've hit the hard limits
> > in ext3.
>
> Ahh... _that_ said, it looks like (until they fix it) XFS is the best
> choice for punishing your enemies with. :o
>
> I hate to say it, but this is not the only place I have heard Linux/XFS
> horror stories. Of course I actually love experimenting, and there's
> nothing wrong with a work in progress, just so long as it's labeled.
>
> I guess the moral of the story is that if you've got a big partition, I
> hope you've got an even bigger backup tape. ;)
Maybe the moral is you should use what everybody else uses. ext2/3 and
to some extent Reiser are very well tested because practically everybody
uses them. Major bugs in ext3 are readily apparent because it has
millions of users. Major bugs in XFS are not found until one of sgi's
17 customers happens to trip over one. Worse, XFS is the sole user of
lots of in-kernel code. ext3 uses lots of underlying code that is also
used by other kernel pieces. The segmentation helps in rooting out
problems.
It's a many-eyes problem. There's a "Safe Operating Area" for most
software. Probably if you went off and made a 4TB ext3 volume with the
journal on a iSCSI device and then scattered it with 150 million files,
you would run into a problem. That's unknown territory. But staying
within a normal < 1TB volume with < 10 million files will almost
certainly work without a problem.
Once I worked at a company and we bought NetApp filers. NetApp sales
engineers told us there was no such thing as fsck for WAFL, their
filesystem. 6 months later we encountered something uniquely horrible
known as a "wack". "Oh," they suddenly remembered, "there's no fsck
unless..." We were not a mainstream user. They had never tested our
workload.
We switched to VxFS on EMC. 6 months later VxFS destroyed an extremely
large filesystem. We were out of the mainstream. Never been tested.
We switched to many small Linux servers with medium-sized ext2
partitions. No problems ever.
End of fable :)
-jwb
Reply to: