You are not logged in.

#1 2014-08-17 11:10:22

grandtheftjiujitsu
Member
From: Georgia, USA
Registered: 2013-07-27
Posts: 85
Website

Filesystem Testing [See List]

Preface:  I was getting the itch to do some experimenting and wanted to share my findingings.  I know that Phoronix churns out regular benchmarks of filesystems and many filesystem vs filesystem discussion have taken place here.  However, I wanted to do the testing myself on my particular set up and see what I found. BTRFS and Ext4 have worked well for me in the past so I began said experiments with lesser used filesystems.  For reference, my boot times with Ext4 and BTRFS were usually 21 - 26 seconds.

Tested File Systems:

  • JFS

  • XFS

  • Ext4

  • Btrfs

  • ReiserFS

  • Reiser4

  • Nilfs2

  • F2fs

Process:

  • Two rsync tests were performed in an installation-type environment where an environment was loaded on the specified hardware and used to move data from the source to the target.

  • Boot time

  • The following were conducted in the newly installed / rebooted environment.  I/O Scheduler and mount options are base on recommendations from the Wiki.

    1. Copying a large file (comic-books.tar.bz2) to /dev/null

    2. Uncompressing a kernel (3.16.1)

    3. Comressing (.tar.xz) a kernel (3.16.1)

Results
Google Drive Spreadsheet

These were all pretty short lived installations as I didn't have a specific goal / need in mind going in.  I just wanted to test the waters and see what I found.  Suggestions, discussion, and comments are welcome.

Summary:    Previously I think I had gotten the numbers jumbled, the point is still pretty much the same.  Btrfs is the clear winner.  However, a solid case can be made for XFS under these conditions.  Ext4 was surprisingly luke warm, while JFS seemed to be slacking.  Reiser4 was very slow to boot and seemed to have an issue with /etc/crypttab or multiple encrypted partitions.

Last edited by grandtheftjiujitsu (2014-08-21 03:47:50)

Offline

#2 2014-08-18 18:10:31

Pse
Member
Registered: 2008-03-15
Posts: 383

Re: Filesystem Testing [See List]

Thank you, I've been wanting to see recent results for filesystem benchmarks. Two things I noticed:

- XFS noop results for copying the large file appear to be wrong (i.e. cached). Almost 6GB/s.
- Btrfs is faster for writes, standard FSes are faster for reads (which is to be expected).

Offline

#3 2014-08-18 18:47:24

grandtheftjiujitsu
Member
From: Georgia, USA
Registered: 2013-07-27
Posts: 85
Website

Re: Filesystem Testing [See List]

That's not a typo for XFS (noop).  I noticed it right away and ran the same test a couple more times with similar results.  Though it does seem peculiar, I'm not sure what the issue was.

Offline

#4 2014-08-18 18:58:32

Pse
Member
Registered: 2008-03-15
Posts: 383

Re: Filesystem Testing [See List]

Is there any chance the file was cached in memory? The kernel can be quite smart sometimes when caching data.

Offline

#5 2014-08-18 21:08:15

Jristz
Member
From: America/Santiago
Registered: 2011-06-11
Posts: 930

Re: Filesystem Testing [See List]

what about Nilfs2, F2FS? there are competitor too in linux.

but anyway, look like btrfs have a green horizont...


Well, I suppose that this is somekind of signature, no?

Offline

#6 2014-08-19 05:05:40

grandtheftjiujitsu
Member
From: Georgia, USA
Registered: 2013-07-27
Posts: 85
Website

Re: Filesystem Testing [See List]

@Pse - That must have been it.  The spreadsheet has been updated.

@Jristz I was testing on a HDD so I didn't see the point in using F2FS.  I'll look into Nilfs2.

Offline

#7 2014-08-19 15:20:52

Pse
Member
Registered: 2008-03-15
Posts: 383

Re: Filesystem Testing [See List]

Thanks again for taking the time to do this. A couple of things I noticed during my use of various filesystems:

- Ext4 is really fast when untaring a kernel tarball, however if you untar over a previous tarball it becomes slower (i.e. it appears Ext4 is really fast populating empty directories, not so much when there is data in them). XFS appeared to be better in my tests for this particular case.

- Btrfs is slow for VM like operations. I wrote a little app that opens a file, writes random data to it, then modifies that data at random places, then attempts to read the data back. This is obviously a bad case for Btrfs, due to its CoW nature. XFS/Ext4 where much better, as expected.

Last edited by Pse (2014-08-19 15:21:32)

Offline

#8 2014-08-20 10:24:37

grandtheftjiujitsu
Member
From: Georgia, USA
Registered: 2013-07-27
Posts: 85
Website

Re: Filesystem Testing [See List]

Thanks for the input.  My general consensus is (not too atypically) that Ext4 is still "tried and true", though Btrfs has served me well; this little escapade has got me looking awfully hard at XFS.  I may upgrade to an SSD in the near future and then I may look at the Flash-specific filesystems.

Offline

#9 2014-08-20 23:01:01

Jristz
Member
From: America/Santiago
Registered: 2011-06-11
Posts: 930

Re: Filesystem Testing [See List]

I was lookint at yours test and noticed that JFS was not tested or that look for me) with deadline.
There so many report that the best elevator for JFS is always deadline.

Anyway I wanna see a f2fs in HD test, maybe sound crasy but what is the fun in thing inside the box?


Well, I suppose that this is somekind of signature, no?

Offline

#10 2014-08-21 03:44:04

grandtheftjiujitsu
Member
From: Georgia, USA
Registered: 2013-07-27
Posts: 85
Website

Re: Filesystem Testing [See List]

You got it!  Spreadsheet has been updated with a split section for JFS like that of the XFS tests.  The results weren't anything special.  However, good call on trying f2fs.  It presented some very interesting results.  Namely, its writes speeds were phenomenal.

Offline

Board footer

Powered by FluxBB