XFS, Btrfs, EXT4 Battle It Out On Linux 3.4
Phoronix: XFS, Btrfs, EXT4 Battle It Out On Linux 3.4
Following the Linux 3.4 kernel benchmarks from last week, available now are the results from a three-way file-system comparison using the Linux 3.4 kernel as well as the Linux 3.2 and 3.3 kernels for reference. The three file-systems being pitted against each other are Btrfs, EXT4, and XFS.
It would be interesting test with larger blocks of 64KB.
It would be interesting to know what commits caused these regressions
In fact anything would be more interesting than this article
I'm away to crawl back under my rock
And not surprisingly, EXT kicked ass.
XFS is basically a fringe filesystem.
BTRFS is a high-feature filesystem and no doubt has some performance issues as a result. In some cases, probably a reasonable tradeoff.
These tests are meaningless on (just) SSD drives
XFS and BTRFS can manage Exabyte file systems where in many cases hard disk drives will be used due to their better bank for buck (and maybe longer lifetime) over SSDs.
HDDs have seek and rotational latencies and the file systems make efforts to reduce these latencies in order to improve performance, however a significant number of recent FS comparisons on Phoronix are run exclusively on SSD drives.
While this may reveal something about the tested systems it reveals absolutely nothing about any real life system in which I I would be considering using these file systems.
If there is a need to benchmark on SSDs, at least report parallel results performed on HDDs - as has been in some past FS comparisons on Phoronix.
I actually doubt it has performance issues as a result of having a lot of features. As long as these features aren't in use, I doubt they would have much impact. My theory is that it is just far less optimized than EXT4 or XFS. Both have had years and years of work to get to the point they are at now, BTRFS is way, way behind where maturity is concerned.
Originally Posted by droidhacker
can you do some benchmarks with JFS please ?
btrfs is doing pretty good
there should be a btrfs vs btrfs showdown, as there are interesting choices, for example:have larger nodes/leaves with -n 32k -l 32k , add lzo compression.
it would be interesting to see the impact of these, but it's going to be a winner regardless, look at some of these seekwatcher videos to compare..
Can't you search a little bit?
Originally Posted by fitzie
What about Linus?
Shouldn't Linus Torvalds have run simple benchmarks like this, noticed the regressions, and done something about them before release? Or if not Linus, then.... somebody? I would still think that since Linus is in charge he would have some sort of regression-testing procedure in place, like OpenSUSE's OpenQA automated testing system. Or Phoronix. I just don't understand how the kernel can be so complex and everyone working on it so smart and then there's things like this and the power problems that no one notices and then requite Matthew Garrett to fix after he gets told it's impossible to fix.