Announcement

Collapse
No announcement yet.

Oracle Plans To Bring DTrace To Linux

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

  • Zetbo
    replied
    https://lwn.net/Articles/461660/

    Inside of Oracle, we've decided to make btrfs the default filesystem for
    Oracle Linux. This is going into beta now and we'll increase our usage
    of btrfs in production over the next four to six months. This is a
    really big step forward, but it doesn't cover btrfs in database
    workloads (since we recommend asm for that outside of the filesystem).

    What this means is that absolutely cannot move forward without btrfsck.
    RH, Fujitsu, SUSE and others have spent a huge amount of time on the filesystem
    and it is clearly time to start putting it into customer hands.
    I've been using btrfs like 4 months now and it works fine(for me). I have raid10 on four drives. My root is on ext4, but I am going to change it this year. I don't have critical data, so I don't give a shit if everything goes sideways. It's not ready for serious bussiness use or if you have critical data. For normal desktop usage it's quite fine. As always backups are your best friend

    Leave a comment:


  • kebabbert
    replied
    Originally posted by Yfrwlf View Post
    Stop calling software "technology". Dtrace is a program. You act as if it's something super amazingly special and that it'd be impossible for anyone else to duplicate it (like strace?).

    kthnx
    Ok, "technology" is not the correct word, you are right on this.

    However, everybody is copying or porting Solaris DTrace. IBM AIX has a copy, called ProbeVue. FreeBSD has ported DTrace. Mac OS X has ported DTrace. Linux has several DTrace copies - none of them good as the original. Just like ZFS - everybody wants it and has ported it, or copying it: BTRFS.

    So, yes the copies are immature and inferior. It is difficult to reinvent superior software as ZFS, or DTrace or Containers or CrossBow or SMF or... - because we see that still today, after many years, BTRFS is not functioning well. Ktrace is not functioning well. etc. All the copies are not functioning well. The ports are functioning well: ZFS in FreeBSD and DTrace in Mac OS X - all works well.

    This proves it is difficult to copy Solaris tech, it is so new and unique. It is not just like, a polished version of a software, no they are totally different animal and does not look like anything else on the market. For instance, DTrace was totally unique and revolutionary when it showed up and everybody was baffled. Several Linux devs switched to OpenSolaris just to get DTrace.

    Leave a comment:


  • kebabbert
    replied
    Originally posted by david_lynch View Post
    At one time Solaris zones, like dtrace, was a killer feature. But I've used openvz/PVC for some time and they seem to be as good or better than solaris zones. If oracle expends the effort to port the solaris flavor of containers to their increasingly divergent linux variant, then we'll have at least 5 separate overlapping OS level virtualization solutions on linux. Oracle zones (Ozones?), LXC, Vserver, OpenVZ and PVC. How cool would it be if instead of further fragmenting the picture, Oracle would help out with one of the existing container implementations...
    Most people agree that BTRFS is a ZFS copy, but less mature. Maybe 3-5 years from now, BTRFS will be released as v1.0. But ZFS development has carried on even further, so BTRFS will still lag behind ZFS. Do you also suggest that Oracle should kill ZFS, and help out with ZFS instead? To decrease fragmentation? ZFS is superior to BTRFS, much more mature and used in Enterprise production today. BTRFS is a copy. Why kill the original, and help out with the copy instead?

    The same with Solaris containers. IBM has also copied Solaris containers, and are calling it WPAR(?). If Linux has lot of immature filesystems or lot of immature container implementations - why kill the original and help out with a copy? Wouldn't it be better if Linux killed all immature alpha phase containers and instead helped out with the original superior ZFS and Solaris Containers?

    Leave a comment:


  • Yfrwlf
    replied
    Stop calling software "technology". Dtrace is a program. You act as if it's something super amazingly special and that it'd be impossible for anyone else to duplicate it (like strace?).

    kthnx

    Leave a comment:


  • staalmannen
    replied
    Originally posted by Kelimion View Post
    I doubt that DTrace could be made into a loadable module. Wouldn't it be too invasive in too many areas of the kernel for it not to be?
    They'd have to sprinkle all manner of hooks in their GPL'd kernel which when enabled target something in a single proprietary module. Sure, that's possible, but you'd think that the maintenance upkeep on keeping all of that working and tested version after version to lose them more money than they'd likely gain.

    That and it's hard to see how DTrace alone would make someone move from RHEL to OUL, or that there would be so many of them that it be profitable for Oracle to do this (in the obscured manner Michael alluded to).

    It leaves me to wonder if they're not going to make a shim on top of the current tracing infrastructure that speaks DTrace, to make it easier to push their Solaris clients to Linux, hoping they'll choose OUL instead of RHEL. This would allow Oracle, if btrfs matures and/or a ZFS port follows, to move development resources from Solaris to Linux over time.

    Time will tell.
    There is already a DTrace for Linux as a loadable module.

    Check:
    http://www.crisp.demon.co.uk/blog/

    Leave a comment:


  • david_lynch
    replied
    And Solaris zones too -

    At one time Solaris zones, like dtrace, was a killer feature. But I've used openvz/PVC for some time and they seem to be as good or better than solaris zones. If oracle expends the effort to port the solaris flavor of containers to their increasingly divergent linux variant, then we'll have at least 5 separate overlapping OS level virtualization solutions on linux. Oracle zones (Ozones?), LXC, Vserver, OpenVZ and PVC. How cool would it be if instead of further fragmenting the picture, Oracle would help out with one of the existing container implementations...

    Leave a comment:


  • Cyberax
    replied
    Well, I worked with both STap and DTrace. I can't say that STap is worse than DTrace.

    It's just not used that much. Mostly because dynamic tracing is not required that often - in reality it's mostly a debugging tool.

    Leave a comment:


  • kebabbert
    replied
    Wow!

    This is really really good news! Now Systemtap and all the other crappy copies can rest in piece and Linux can get the original and unique DTrace that every developer wants!

    Leave a comment:


  • Cyberax
    replied
    At this point of time, the only thing required for Linux to achieve theoretical feature-parity with DTrace is to merge uprobes into the main kernel.

    Leave a comment:


  • Kelimion
    replied
    I doubt that DTrace could be made into a loadable module. Wouldn't it be too invasive in too many areas of the kernel for it not to be?
    They'd have to sprinkle all manner of hooks in their GPL'd kernel which when enabled target something in a single proprietary module. Sure, that's possible, but you'd think that the maintenance upkeep on keeping all of that working and tested version after version to lose them more money than they'd likely gain.

    That and it's hard to see how DTrace alone would make someone move from RHEL to OUL, or that there would be so many of them that it be profitable for Oracle to do this (in the obscured manner Michael alluded to).

    It leaves me to wonder if they're not going to make a shim on top of the current tracing infrastructure that speaks DTrace, to make it easier to push their Solaris clients to Linux, hoping they'll choose OUL instead of RHEL. This would allow Oracle, if btrfs matures and/or a ZFS port follows, to move development resources from Solaris to Linux over time.

    Time will tell.

    Leave a comment:

Working...
X