James G. Sack (jim) wrote:
from LWN, a link to
Jeffrey W. Baker: ZFS, XFS, and EXT4 compared. (September 5, 2007)
http://lwn.net/Articles/247811/
which links to the actual data, but I will quote his summary:
"""
Short version: ext4 is awesome. zfs has absurdly fast metadata
operations but falls apart on sequential transfer.
That's too strong a statement given the data. Or, at least, he should
say, "ext4 falls apart on metadata operations" given that the gap is
about the same in reverse.
I'm going to completely ignore the hardware RAID specs. They are
uninteresting, IMO. If I'm using ZFS, I'm using it to *avoid* hardware
RAID.
Comparing the software implementations, ZFS holds at about a factor of 2
slower than ext4 (is ext4 checksumming data like ZFS?) for sequential
operations. ext4 holds at about a factor of 2 slower than ZFS for
random operations.
And ZFS kicks the stuffing out of xfs for the Postmark benchmark from
NetApp (factors of *5* in places).
The multithreaded data is nice, but I generally worry about those kind
of benchmarks since they often are limited by things other than the
actual performance of the filesystem.
So, in short, ZFS is optimized for random operations; ext4 for
sequential ones.
Given that Sun's customers tend to be database folks, that's not
unsurprising.
So, if you're compiling lots of stuff, you want ZFS. If you are doing
lots of video editing, you probably want EXT4.
Personally, I'll take ZFS. It matches my usage patterns better *and*
actually gets compliance tested by Sun.
As always, YMMV,
-a
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
