With the development of log-structured filesystems and the introduction of (relatively) inexpensive SSDs, is there a best choice for a log-structured filesystem for use on Linux with SSD drives?
1
-
Are you talking about the multi-cell or single-cell SSDs (X25-M vs X25-E from Intel, for example)?– jharleyMay 28, 2009 at 19:46
Add a comment
|
6 Answers
NILFS2 has been merged into the current development kernel, and will be available in kernel 2.6.30 (You can run a -rc kernel if you're impatient). It apparently performs very well on SSDs.
As with anything recently merged into the kernel, you'll have to make your own decision regarding stability and maturity.
It seems to be that the answer is to use a filesystem you would use with normal disk.
Linus seems to second this idea.
See Theodore Ts'o 's article regarding this.
In his other blog entry he writes how to align to ssd's erase block size.
It seems to me that modern SSD's are smart enougth that you should not care. My choice of filesystem is XFS btw.
There are two mount options: relatime, noatime which can be used to reduce disk writes at the expense of modified posix semantics. Reduced disk writes are considered as a good idea with SSD's Theodore Ts'o made some bechmarks regarding the issue.
-
1Yes, modern SSDs targeted for server/desktop use spread writes quite fine. Just make sure that you reduce the amount of "unncessary" writes, if you're concerned about device wear. Disabling atime updates, as suggested above, is a good first start if you don't run applications which depend on it's presence. Jun 4, 2009 at 17:26
-
-
I appreciate the answer, but I was specifically asking about log-structured file systems. Upvote for the valuable info, though. Jul 6, 2009 at 17:41
-
+1 for XFS, it seems to be much better than EXT3 when it comes to journaling large blocks of data (wich means less tear of the SSD).– pauskaJul 20, 2009 at 14:39
I have seen it suggested that the journalling of ext3 is neither desirable (imposes extra writes) nor particularly useful (writes are fast anyway) on Flash drives, and that thus ext2 is to be preferred of the two.
-
1
-
1google.com/search?q=ext3+journal+flash+ssd gives plenty of hits (as it did when I was preparing to install on an SSD a few months back) but many of them appear to be reference-less just like my answer :-( However thunk.org/tytso/blog/2009/03/01/… (second hit when I searched just now) appears to bust this myth, showing that mount option "noatime" is a huge win compared to losing the journal. Live and learn.– mlpMay 28, 2009 at 9:19
Consider YAFFS
-
-
because it might be 'a best choice for a log-structured filesystem for use on Linux with SSD drives' - the question doesn't give a lot of detail to qualify 'best', it is an active LSFS for Linux that has its supporters but your mileage may vary. I don't see why you'd want to 'down' a perfectly decent, but brief, suggestion that to an equally brief question - especially when I even linked to where you can find out a lot more about the subject - seems rather ungrateful and ungracious.– Chopper3May 8, 2009 at 11:04
-
3YAFFS is designed specifically for NAND flash memory devices. You should not be confusing raw flash memory with packaged flash based "disks". The former is generally accessed through the specialized mtd driver, while the latter would present itself as a generic block device. The SSDs that McWafflestix talk about are just regular disks, block devices, as far as the OS is concerned. It internally takes care of issues such as wear leveling so that you can happily run for example ext3 on them.– connyMay 21, 2009 at 11:38
Honestly there is no mature filesystem that is optimized for SSDs. Use whatever filesystem you would have otherwise.
