On Wednesday 23 October 2013, Greg Freemyer wrote:
On Wed, Oct 23, 2013 at 8:52 AM, Ruediger Meier <sweet_f_a@gmx.de> wrote:
On Wednesday 23 October 2013, Greg Freemyer wrote:
"Carlos E. R." <robin.listas@telefonica.net> wrote:
On 2013-10-23 12:59, David T-G wrote:
Any thoughts, shy of religious warfare, on the best FS to use for a
data
and archive server? [I promise this is it for me for today ;-]
XFS?
10 years ago I researched the same question and chose xfs. I've followed both xfs and ext3/ext4 development since then and would still pick xfs of the 2 (for the described workload).
XFS is _extremely_ slow when creating/deleting files, AFAIR something about factor 100! worse than the ext family. That's why I would not used it for (rsnapshot) backups and most other use cases.
I think you recall from 2003 not 2013!
No, from 2011, kernel 2.6.37. Even I re-validated my results using 3.0.80-52, which is the up-to-date kernel of SLE11. I consider it my own huge mistake that I've started using xfs at all without benchmarking before. It was a lot work to migrate back and there are still some xfs partitions left which are as slow as measured in 2011.
As 18 months ago, the delta was almost gone. At that time Dave Chinner (one of the core XFS developers) was argueing that he only saw a multi-year future in XFS and BtrFS. As such his argument was people should move to XFS if they wanted short term stability because it was going to be around and well supported for years.
It sounds right somehow, but xfs was still 100 times slower probably on all existing distros at this time, just 18 months ago. So I wouldn't have recommended it at that time. And even now it will be too slow on most existing enterprise distros.
Then for desktops and small servers, migrate them to BtrFS when it was ready. I'll let you read some of it yourself:
Planning migration to btrfs in future is even one more argument for ext2/3/4 because btrfs-convert does not work for xfs.
From <http://lwn.net/Articles/476565/> Dave Chinner says:
It's good to know, maybe I will keep my last existing XFS partions and try a recent kernel.
[...] XFS is now so close ext4 performance on single threaded metadata intensive workloads that it ext4 has lost the one historical advantage it held over XFS.
Hm, maybe this performance difference is really gone now. But it was such a huge difference (feels like comparing HD with floppy disk) ... and it was not fixed for many many years ... so at least for me it will take some time before forgetting the bad taste of xfs. cu, Rudi -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org