Announcement

Collapse
No announcement yet.

Lead Btrfs FIle-System Developers Join Facebook

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

  • Lead Btrfs FIle-System Developers Join Facebook

    Phoronix: Lead Btrfs FIle-System Developers Join Facebook

    Chris Mason and another lead Btrfs developer have sent out a kernel patch today updating their email addresses as they will be joining Facebook to work on the Linux file-system...

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

  • #2
    It's been in development for like 5 years already and still no final format, no stable release. It's one of those projects I stopped caring about.

    Comment


    • #3
      Originally posted by mark45 View Post
      It's been in development for like 5 years already and still no final format, no stable release. It's one of those projects I stopped caring about.
      You cared enough to read this and post a comment.

      Comment


      • #4
        Originally posted by mark45 View Post
        It's been in development for like 5 years already and still no final format, no stable release. It's one of those projects I stopped caring about.
        Since you commented, I think you care. You're just frustrated like the rest of us. Let's hope Facebook can cram more manhours out of these guys. They probably have deeper pockets than Fusion-IO. Maybe it isn't manhours that's bottlenecking the Btrfs progress though.

        Comment


        • #5
          Jeez, can't these guys keep a job? ;D

          Comment


          • #6
            Designing a filesystem of this level of complexity is obviously hard. I mean, really hard. And that's why there are not much hackers around there which can work on this sort of stuff. We should accept that this sort of project take times to mature.

            That being said, I feel like there is an issue in the way btrfs projects is conducted. But my opinion should be taken which a lot of skepticism, since i'm not really qualified to talk neither about btrfs himself nor about its development.

            I feel like they are not doing it in the right order. It's true that to fix a disk layout, memory structures and core algorithms, you need to design it as a whole, thinking about the feature you will implement. But adding features like compressions before having a fsck tool does not look rights. Maybe it would have been better to stabilized a core set of features, and start adding new ones like raid, compression, dedup, etc after this stabilization pass.

            Comment


            • #7
              Facebook, that's interesting. There's a division of Facebook working with the D language, so Facebook funds a portion of its development, too. In case of D, they use it as a framework for code testability and stability. I wonder what their interest in Btrfs is... Probably reliable data storage. Overall the change probably doesn't mean much, as Fusion-io is a fairly large company as well.

              enjolras, no, the priorities are quite fine. Btrfs doesn't need a fsck, because it does all the checking and repair at runtime (which can also be manually forced with scrub), while recovery is done via mount options. There is little reason to also make an offline tool for it.

              Comment


              • #8
                Originally posted by enjolras View Post
                I feel like they are not doing it in the right order. It's true that to fix a disk layout, memory structures and core algorithms, you need to design it as a whole, thinking about the feature you will implement. But adding features like compressions before having a fsck tool does not look rights. Maybe it would have been better to stabilized a core set of features, and start adding new ones like raid, compression, dedup, etc after this stabilization pass.
                Are you sure they plan develop a fsck tool at all? I don't think zfs has it? I thought it was unnecessary for this sort of file system?

                Comment


                • #9
                  Originally posted by Akka View Post
                  Are you sure they plan develop a fsck tool at all? I don't think zfs has it? I thought it was unnecessary for this sort of file system?
                  Ideally Btrfs will either have the old data, or the new data. Never a middle issue resulting in corruption. BUT thats ideally and in theory, so you could run into situations where everything just gets completely hosed either due to FS bugs or completely random uncontrollable outside factors

                  Comment


                  • #10
                    Originally posted by GreatEmerald View Post
                    Facebook, that's interesting. [] I wonder what their interest in Btrfs is... Probably reliable data storage.
                    Yes i first thought: Facebook?!?!
                    But they have to store a lot of data so i guess they want to speed that development a bit. Or push that into a direction that suits them best.

                    Comment


                    • #11
                      As far as I understand, you always need an fsck tool. You can design a COW FS to make sure that if a OS crash or a power failure happens in the middle of a write, you can restore back to the last flushed coherent state.

                      But that only deals with OS crashs or power failures. If you have a disk error for instance, and you manage to extract the data out of the disk, you may want to rebuild a coherent state which omits the corrupted parts. But to be honest, I don't know how btrfs does that or if it does. I've been using btrfs for 6month and came back to ext4 due to the silly metada allocation space full which did not reported the FS as being full and prevented me from booting my OS three times after I upgrade the distribution. Believe, when you have no internet access and you tries to fix that from a busybox emergency environment, it drives you crazy not to even be able to issue "rm" because you get ENOSPC. I got angry and reformatted all my volume, even if i know that this issue has been addressed since then.

                      Anyway, fsck was just an example for my point, the compression or dedup cases still apply.

                      Comment


                      • #12
                        Originally posted by mark45 View Post
                        It's been in development for like 5 years already and still no final format, no stable release. It's one of those projects I stopped caring about.
                        Try 6 years. Development began in late 2007.

                        Compare to ZFS. Development started in 2001. Then ZFS was included in the June 2006 update of Solaris 10. That is 5 years from beginning of development to deployment in a production OS.

                        But ZFS pioneered a lot of new filesystem concepts. btrfs already had all of those ideas available to build on in 2007, so btrfs should have been ready faster than ZFS. Implementation should be faster when you do not need to invent the wheel (only reinvent it).

                        Instead, after 6 years of development, btrfs is still not ready to be the default root filesystem of a stable OS release.

                        Unfortunately, I think it is evident that the project management of btrfs is ineffective. Announcements like this that the project management is not changing may be meant to be reassuring, but I am not reassured.
                        Last edited by jwilliams; 12-04-2013, 05:51 PM.

                        Comment


                        • #13
                          Why is it news that a developer opens a Facebook acco... oh wait.

                          Comment


                          • #14
                            Originally posted by jwilliams View Post
                            Try 6 years. Development began in late 2007.
                            It's almost 7 years old

                            commit be0e5c097fc206b863ce9fe6b3cfd6974b0110f4
                            Author: Chris Mason <chris.mason@oracle.com>
                            Date: Fri Jan 26 15:51:26 2007 -0500

                            Btrfs: Initial checkin, basic working tree code

                            Signed-off-by: Chris Mason <chris.mason@oracle.com>


                            https://lkml.org/lkml/2007/6/12/242

                            "After the last FS summit, I started working on a new filesystem that
                            maintains checksums of all file data and metadata."

                            I think that he is talking about 2006 fs summit, because 2007 was at February 1213
                            https://www.usenix.org/legacy/event/...p/lsf07cfp.pdf

                            Comment


                            • #15
                              I guess Facebook really wants to port over to Btrfs from ext4

                              Comment

                              Working...
                              X