Announcement

Collapse
No announcement yet.

AFS File-System Driver Overhauled For Linux 4.15

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

  • AFS File-System Driver Overhauled For Linux 4.15

    Phoronix: AFS File-System Driver Overhauled For Linux 4.15

    Red Hat developers have been working on an overhaul of the AFS file-system's kernel driver for the just-opened Linux 4.15 kernel merge window...

    http://www.phoronix.com/scan.php?pag...led-Linux-4.15

  • #2
    That's a blast from the past. When I was a student at CMU I actually jumped through the flaming hoops that were required to get my Linux box talking to my AFS share on the University network (a whopping 10 MEGABytes, which was incredibly puny even at that time since AFS wasn't being very well maintained). That included plenty of weird kernel & userspace stuff plus making sure I was actually authenticating via Kerberos properly. This was around the time that the 2.6 Kernel was going mainstream, so it's been a while.

    AFS and its stillborn successor Coda have a bunch of crazy features that are pretty cool sounding, but the complexity to actually get the system working has meant that it has had very limited acceptance.

    Comment


    • #3
      I'm curious if Red Hat's spiked interest in AFS has anything to do with their new Stratis stack.
      Does anyone have any insight?

      Comment


      • #4
        The in Kernel AFS driver isn't that great it never has been as I understand it even with these changes it is missing alot of stuff that openafs supports much less auristor.

        https://www.auristor.com/ is where it's at these days... Jeff Altman did a ton of work to make openafs modern, but no one wanted to pay him so he just forked it off as auristor I guess. He even had stuff like mobile support working so you could use it over wifi without issues unlike AFS. I met Jeff at UNC-Charlotte and was in some of the meetings with him and the IT department talking about the issues with openafs and kerberos a few years back. (as a student)

        Comment


        • #5
          Why Red Hat needs 3rd distributed FS along Ceph and Gluster? I think they should standardize on one, and divest the other to say Canonical or SuSE. And they get involved with 3rd.

          Comment


          • #6
            Originally posted by doublez13 View Post
            I'm curious if Red Hat's spiked interest in AFS has anything to do with their new Stratis stack.
            Does anyone have any insight?
            The kernel AFS and AF_RXRPC work is a personal project of David Howells which AuriStor, Inc. is supporting.

            Comment


            • #7
              Originally posted by cb88 View Post
              The in Kernel AFS driver isn't that great it never has been as I understand it even with these changes it is missing alot of stuff that openafs supports much less auristor. https://www.auristor.com/ is where it's at these days... Jeff Altman did a ton of work to make openafs modern, but no one wanted to pay him so he just forked it off as auristor I guess. He even had stuff like mobile support working so you could use it over wifi without issues unlike AFS. I met Jeff at UNC-Charlotte and was in some of the meetings with him and the IT department talking about the issues with openafs and kerberos a few years back. (as a student)
              kAFS has come a long way in the last 18 months which significant improvements since the kAFS Hackathon at Vault [url]http://events.linuxfoundation.org/events/vault/extend-the-experience/kafs-hackathon{/url]. The project page https://www.infradead.org/~dhowells/kafs/ has not been fully updated to reflect the changes that are included in the Linux 4.15 merge window.

              The primary weakness of OpenAFS and AuriStorFS on Linux are the fact that both are out of tree file system kernel modules that are non-GPL. As a result they cannot make use of all of the Linux kernel's vfs functionality. Being out of tree, it is always a catch up game tracking mainline Linux kernel development and then reacting after the fact to each distribution's choice of backports. Being in-tree kAFS will eventually become the basis upon which Linux containers built upon the global /afs file namespace will become portable and mainstream.

              kAFS now supports all of the IBM AFS 3.6 and OpenAFS RXAFS and RXAFSCB RPCs except for the metadata information leaking debugging and statistics opcodes.

              kAFS supports IPv6 which OpenAFS does not; and it implements some of the AuriStorFS services so that IPv6 can be used not only with the Location service but with file services as well.

              The AF_RXRPC socket class is independent of kAFS and can be used as the basis for lightweight IoT applications.

              The kAFS future is bright. AuriStor will always provide its own AuriStorFS kernel module which can be installed side-by-side with kAFS. However, we view kAFS as critical to the success of AuriStorFS. Only by ensuring that every Linux device can access /afs can AuriStorFS be competitive with NFS and CIFS solutions.

              AuriStorFS (formerly YFS) is commercial because it was funded under U.S. Dept of Energy SBIR grants https://www.sbir.gov/sbirsearch/detail/354025 with the unanimous support of the OpenAFS Elders.

              Comment


              • #8
                kafs has been enabled in Fedora Rawhide for the 4.19 kernel. When the 4.19 kernel is pushed to future Fedora releases kafs will be available to end users.

                Comment

                Working...
                X