If you are asked to list the types of filesystems supported in Linux, I am sure you will be able to list quite a few of them. But when it comes to listing the pros and cons with respect to the performance of these filesystems in Linux, one is not so sure right? This may be because not many have benchmarked these filesystems and published their findings online.Now things have changed though, because Justin Piszcz has written a two part series listing his findings regarding how each filesystem fared performance wise under Linux. He has published his findings for Linux kernel 2.4 as well as 2.6 .
The filesystems that were included in the benchmark were EXT2, EXT3, JFS, ReiserFS v3, ReiserFS v4 and XFS . To arrive at a conclusion, he performed 17 different tests which included creating 1000's of files on the filesystem and cat'ing a huge file (1 GB) to /dev/null among other things.
So what is the verdict ?
The author has concluded that XFS is the best filesystem in terms of performance and scalability though JFS has improved in some of the tests. And the surprise of all, ReiserFS is the slowest in most of the tests. But I think it would be prudent to read the article first and arrive at your own conclusions.
3 comments:
and you what fs are using ?
personally I always go with an ext3 /boot and a reiser / .... but they have always been laptop or personal machines ....
I usually use reiserfs, simply because it was the first I really used, it works (I imagine though just as well as the others) and because I know how to recover deleted files in it. I once did a recursive rm on / and then tried to see how much of the system could be saved using a livecd.
# reiserfsck --rebuilt-tree -S -l /dev/hda#
Where # of course is the partition number you're trying to check. This recovered about 75-80% of my files. I haven't gone out of my way to learn too much more, so I can't say I know how good the others are. :o)
A number of filesystems tested (think: ReiserFS, Reiser4) were crippled by the bizarrely underpowered CPU (500 MHz).
On a more modern/realistic setup the test results would have been rather different.
The tests are good for what they are; but billing them as "comparisons of various filesystems under linux" implies that the test system is a reasonably current machines. It isn't. It's a 500 MHz sloth with 200 GB drives that became affordable months or years after CPUs of that speed stopped being produced.
Post a Comment