Re: (A MUST READ!!!!!) The Ultimately Fastest Ubuntu Desktop/Server You Can Have!
A more recent benchmark set (though all micro-benchmarks should be taken with a fairly large grain of salt) shows that ext3, in metadata journaling mode (data=writeback), performs competitively with XFS under a range of typical server loads (using [no]atime or APC apparently made little difference on either FS; see stats). Scroll to the bottom for the raw stats pdf file.
Similarly, a benchmark set from a few years ago, showed that on average, taking the best and weakest points of each system, they basically are all the same. One may excel in some job that another lags in, while the other beats the first in a different area. Or at least the numbers are close enough, on average, that they are not above the signal:noise threshold. The article also shows how to do some nice speed tweaks on XFS (which is its main purpose, BTW, the benchmarks are kind of incidental).
Ps. Just to clear up a point that someone said earlier: XFS doesn't "loose" data when the journal is not cleanly flushed. ext3, in full journal mode, gives you the last inodes that were journaled as file fragments (ordered journal mode is more complicated). XFS gives you the metadata only (which points to extents with a size but no data). When ext3 is in metadata journal mode, you get the same behavior -- XFS just had a bug that caused it to write nulls ("zeros") rather than freeing the extents. It is interesting to read SGI's FAQ about XFS, starting with the Q: "Why do I see binary NULLS in some files after recovery when I unplugged the power?"
All my partitions are ext3, BTW -- and hopefully ext4 soon, like the previous poster said! :)
Last edited by MonkeeSage; April 11th, 2008 at 02:35 AM.
[People] are usually satisfied with bad argument only when their convictions rest on other grounds. (John Oman, Grace and Personality [New York: Macmillan, 1925], p. 38).