Announcement

Collapse
No announcement yet.

Running ZFS On The Linux 4.1 Kernel

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Running ZFS On The Linux 4.1 Kernel

    Phoronix: Running ZFS On The Linux 4.1 Kernel

    With the recently released ZFS On Linux 0.6.4.2 there is added support for the Linux 4.1 kernel. After carrying out the recent 6-way file-system comparison on Linux 4.1 I decided to run some fresh tests of this popular, out-of-tree file-system.

    http://www.phoronix.com/vr.php?view=21794

  • #2
    It would be interesting to see if there is a difference in cpu usage (and thus maybe power consumption?) and ram usage in these file systems and how they perform in cpu/ram restricted environments.

    Comment


    • #3
      Huh? Why the bad database and multi-threading results? I always though ZFS was designed for that purpose.

      Comment


      • #4
        Originally posted by oleid View Post
        Huh? Why the bad database and multi-threading results? I always though ZFS was designed for that purpose.
        ZFS default recordsize is 128K as far as I remember. A database uses 8k oder 16k. You have to configure ZFS for the database of your choice. If you just use the default values you wasting performance.

        So this benchmark is pretty useless.

        ZFS is _NOT_ a "fire and forget" filesystem.

        Comment


        • #5
          Originally posted by -MacNuke- View Post
          ZFS is _NOT_ a "fire and forget" filesystem.
          No. And for a real high-performance database server, you would typically also be adding ZIL and L2ARC caches to your pool, plus running over a number of data discs to get more disc bandwidth (as well as redundancy).

          Comment


          • #6
            Originally posted by Namenlos View Post
            It would be interesting to see if there is a difference in cpu usage (and thus maybe power consumption?) and ram usage in these file systems and how they perform in cpu/ram restricted environments.
            Actually I would love to see that as a standard test.
            Run the same test with 64, 128,256,512 and 1024MB of RAM. That would definitly show some interesting figures.
            And 1...16 cores, to see if multithreading across numa nodes hit a bottleneck.
            And downclocking the CPU to very low, or have some other means to get a nice graph out of throughput, fairness (I've seen a lot of starvation in a fast I/O system), cpu usage, and memory usage.
            If you only have 512MB ram, would you even consider btrfs? Or zfs for that matter.
            If you have a SOHO nasan (I use my nas as a san: FCOE with a local ssd for bcache, and it works perfectly), would you serve files, or would you serve blocks. Would you backup to tahoe-lafs, or would you rather go ceph. Rsync as backup to what filesystem?

            Comment


            • #7
              Originally posted by oleid View Post
              Huh? Why the bad database and multi-threading results? I always though ZFS was designed for that purpose.
              ZFS really isn't designed for your typical desktop user who just uses things out of the box.

              It's for enterprise use, where you have multiple disks and an admin tweaking the config settings to work with the hardware available and the software using it.

              For that reason, I'm not sure it makes much sense for Michael to benchmark it. He'd have to go to WAY more effort than normal to set it up properly - by which i mean at the very least doing some investigation to see how it should be setup, and then explaining all that in the article. And then he's just going to get a bunch of complaints about how he's doing it wrong, and instead he should tweak some other settings, etc.
              Last edited by smitty3268; 07-07-2015, 02:55 PM.

              Comment


              • #8
                Originally posted by smitty3268 View Post
                For that reason, I'm not sure it makes much sense for Michael to benchmark it. He'd have to go to WAY more effort than normal to set it up properly - by which i mean at the very least doing some investigation to see how it should be setup, and then explaining all that in the article. And then he's just going to get a bunch of complaints about how he's doing it wrong, and instead he should tweak some other settings, etc.
                Research? Effort? Journalism? I think we know what this calls for.... EricG

                Also I'm rather curious how FreeBSD compares given that the linux implementation isn't exactly ideal.

                Comment


                • #9
                  Originally posted by Luke_Wolf View Post
                  Also I'm rather curious how FreeBSD compares given that the linux implementation isn't exactly ideal.
                  Also compare it to the Solaris implementation

                  Comment


                  • #10
                    Testing zfs on a single disk, is like testing a car after removing all tires.

                    Comment

                    Working...
                    X