Btrfs Is Not Yet The Performance King

[CheetoPet]

wow

I find it impressive that such an immature filesystem is so competitive with the others that have been around for way longer. Sure it scored last in most of the benchmarks, but at least its in the same ballpark. I know there have been a lot of performance improvements that have gone into the .30 rc's, so it'll be interesting to see how that kernel improves over .29

[deanjo]

Here is a bit more in depth look at btrfs:

http://www.linux-mag.com/id/7308/1/

As you can see the mount options and compression can make a huge difference.

[Ant P.]

Anyone know what that weird triangle in the bootchart graph is caused by?

[CheetoPet]

Here is a bit more in depth look at btrfs:

http://www.linux-mag.com/id/7308/1/

As you can see the mount options and compression can make a huge difference.

Theres also been discussion on the mailing list about getting a userspace file deduplication tool working. That would be pretty amazing.

[energyman]

how do ext3/4 deal with small files? how is btrfs with thousands of small files? and does btrfs hates user data as much as ext4 does?

reiser4 never led me down. It cares about data. It is fast with all kind of files. I would love to see tests covering it. Especially with lzo or gzip compression turned on.

[Craig73]

I agree that more info is needed before making a judgement
- the linux mag tests show that compression in certain setups can greatly improve performance
- tests should include general desktop usage as well as database tests, compiling tests, and perhaps VM tests as well... stress it out a bit
- comparison of efficiencies of disk space usage (with or without compression). You stated how you allocated the drive - how much free space was there for each file system? After copying 50GB of games, movies, photo's, documents, etc., how much free space is there? After turning on compression, how much free space is there? [and this is performance compression / not archiving compression - but it still saves space]
- include CPU/Memory usage for each test. IE - does managing the file system require more CPU, especially when compression is turned on. IE - where are the tradeoffs.

And btw - I disagree with those who feel the features are un-interesting to desktop users. Online check, defrag, and compression. Data Integrity. Anything to help ensure my data is safe and my system is peforming well is quite significant to me.

[now perhaps other file systems provide the same features at less cost, that is another discussion]

[kraftman]

Here is a bit more in depth look at btrfs:

http://www.linux-mag.com/id/7308/1/

As you can see the mount options and compression can make a huge difference.

It seems btrfs is very flexible.

@Energyman

I don't know about reiser4, but reiserfs will probably gain quite good performance improvement:

http://lkml.org/lkml/2009/5/1/8

And btw - I disagree with those who feel the features are un-interesting to desktop users. Online check, defrag, and compression. Data Integrity. Anything to help ensure my data is safe and my system is peforming well is quite significant to me.

Yeah, I wonder what features those desktop users expect? Cheats in games?

[quintesse]

NILFS?

I would have liked seeing NILFS (recently included in the kernel) added to the list of tested file systems. Up to now I have read very little about in comparison to others but it sounds very interesting.

[Wulfsdale]

I find it impressive that such an immature filesystem is so competitive with the others that have been around for way longer. Sure it scored last in most of the benchmarks, but at least its in the same ballpark. I know there have been a lot of performance improvements that have gone into the .30 rc's, so it'll be interesting to see how that kernel improves over .29

Agree. Butter FS shows great promise in an early stage. Would like to see this test in a year from now. Scalability is a key feature for me.

W

[Phlogiston]

Could you include nilfs next time?