Announcement

Collapse
No announcement yet.

F2FS Gets Improved Read/Write Performance

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

  • F2FS Gets Improved Read/Write Performance

    Phoronix: F2FS Gets Improved Read/Write Performance

    The Samsung-developed Flash-Friendly File-System (F2FS) will exhibit improved read/write performance for some workloads with the Linux 3.14 kernel...

    http://www.phoronix.com/vr.php?view=MTU3OTk

  • #2
    The 8-way comparison you linked to does NOT contain F2FS.

    Comment


    • #3
      Still missing some critical bits

      Namely ACLs and extrended attributes.

      Comment


      • #4
        Originally posted by phoronix View Post
        Phoronix: F2FS Gets Improved Read/Write Performance

        The Samsung-developed Flash-Friendly File-System (F2FS) will exhibit improved read/write performance for some workloads with the Linux 3.14 kernel...

        http://www.phoronix.com/vr.php?view=MTU3OTk
        Is there any forum or thread where someone can go to talk about, discuss and throw in f2fs requests? I would like to show my thanks and offer my time for testing f2fs, but I do not know who to contact or where to post these questions. Thanks

        Comment


        • #5
          Originally posted by SkOrPn View Post
          Is there any forum or thread where someone can go to talk about, discuss and throw in f2fs requests? I would like to show my thanks and offer my time for testing f2fs, but I do not know who to contact or where to post these questions. Thanks
          Samsung seem to be playing f2fs development quite close to their chests.

          My own tests show it as being amazingly stable. I've yet to manage to cause any data loss or FS damage no matter how badly I treat the OS/devices - and that includes unplugging them mid-write as well as powering the entire system off.

          Comment


          • #6
            Originally posted by stoatwblr View Post
            Samsung seem to be playing f2fs development quite close to their chests.

            My own tests show it as being amazingly stable. I've yet to manage to cause any data loss or FS damage no matter how badly I treat the OS/devices - and that includes unplugging them mid-write as well as powering the entire system off.

            That is very good news. I too have been using it but only on my Nexus 7 Android device. It has brought new life into this 2012 tablet to say the least. Makes me wonder about desktop computers and if there will be any noticeable benefits for them.

            Comment


            • #7
              Originally posted by SkOrPn View Post
              That is very good news. I too have been using it but only on my Nexus 7 Android device. It has brought new life into this 2012 tablet to say the least. Makes me wonder about desktop computers and if there will be any noticeable benefits for them.
              I was testing it on desktop systems before deploying on androids (There's a lot of hoop jumping to make it mount on Samsung S2 as a f'instance) and that's where most of the abuse tests took place.

              Performance is on par with ext4 with standard (metadata only) journalling and outruns ext4 with full journalling (which is what COW effectively is).

              It _doesn't_ (yet) provide user_xattr or acls, so if you need those you're out of luck (fscache needs this), and grub doesn't understand it (which means you can't put it on /boot) but in all other cases except ZFS, I'm using it as the default SSD fs.

              Note that the standard drive-info stuff doesn't understand it either, so you can't use LABEL/UUID= in fstab, but if you use /dev/disk/by-id/ then you'll be safe. (referring directly to /dev/sd* runs the risk that the drives may reorder themselves if you boot with a USB stick or rearrange your cabling)

              Comment


              • #8
                Originally posted by stoatwblr View Post
                I was testing it on desktop systems before deploying on androids (There's a lot of hoop jumping to make it mount on Samsung S2 as a f'instance) and that's where most of the abuse tests took place.

                Performance is on par with ext4 with standard (metadata only) journalling and outruns ext4 with full journalling (which is what COW effectively is).

                It _doesn't_ (yet) provide user_xattr or acls, so if you need those you're out of luck (fscache needs this), and grub doesn't understand it (which means you can't put it on /boot) but in all other cases except ZFS, I'm using it as the default SSD fs.

                Note that the standard drive-info stuff doesn't understand it either, so you can't use LABEL/UUID= in fstab, but if you use /dev/disk/by-id/ then you'll be safe. (referring directly to /dev/sd* runs the risk that the drives may reorder themselves if you boot with a USB stick or rearrange your cabling)
                Do you think this fs has a chance to become the go to fs for booting Linux and Windows OS's someday? I know grub should eventually adopt it but with Windows driver support on the road map maybe even use it for Windows 9? I have clients with new computers and very old computers and would love a free performance boost for them from a simple concept such as a filing system. This could help even 10 year old laptops run more modern operating systems possibly. I guess we'll see...

                Comment


                • #9
                  Originally posted by SkOrPn View Post
                  Do you think this fs has a chance to become the go to fs for booting Linux and Windows OS's someday? I know grub should eventually adopt it but with Windows driver support on the road map maybe even use it for Windows 9? I have clients with new computers and very old computers and would love a free performance boost for them from a simple concept such as a filing system. This could help even 10 year old laptops run more modern operating systems possibly. I guess we'll see...
                  Honestly? I hope not and that's despite using it as primary FS on my laptops.

                  F2FS has its uses, but it's primarily designed for low-end storage devices in portable equipment which doesn't have onboard optimisation (sd cards, etc) and as a byproduct it works well in single drive SSD setups.

                  On any system large enough to take multiple devices, ZFS and friends are the way forward if the CDDL vs GPL issue can be resolved amicably.

                  As far as use on windows goes, unix FSes lack a number of semantics which NT requires. That's one of the reasons why NT support for other FSes is lacking.

                  NTFS itself is fairly good, if treated well.

                  The best speedups achievable in older systems usually comes from adding more ram - and specifying enough ram in current ones, bearing in mind that "unused" ram in BSD/Linux/Windows is used as disk cache (We're putting 16-32Gb in scientific desktop systems at the moment, 16Gb in office systems. Complaints about poor performance are mostly a thing of the past)

                  "Standard" windows installations are generally woefully underspecced memory-wise. As soon as you have to touch disk you only get to run at the speed of the disk, so the best way of improving performance is to reduce that to the absolute minimum you can.


                  (Given a choice between processor speed and more ram, take the ram every time. Given a choice between processor speed and faster ram, take the faster ram, it makes a lot more difference)

                  Comment

                  Working...
                  X