Slashdot Mirror

← Back to Stories (view on slashdot.org)

Linux 4.18 Preparing Many New Features While Dropping 100k+ Lines of Code (phoronix.com)

Posted by msmash on from the up-next dept.

An anonymous reader writes: Linux 4.18 development is going strong with recent 4.18-rc1 release. This kernel cycle has dropped 107,210 lines of code so far but Linux 4.18 is adding many new features. The kernel is coming in lighter as a result of the LustreFS code being removed and other code cleanups. On the feature front, Phoronix reports, "ew AMDGPU support improvements, mainlining of the V3D DRM driver, initial open-source work on NVIDIA Volta GV100 hardware, merging of the Valve Steam Controller kernel driver, merging of the BPFILTER framework, ARM Spectre mitigation work, Speck file-system encryption support, removal of the Lustre file-system, the exciting restartable sequences system call was merged, the new DM writecache target, and much more."

13 of 105 comments (clear)

  1. Lustre dead? by afidel · · Score: 3, Interesting

    With Intel shutting down their commercial support business last year and now LustreFS being removed from the mainline kernel is Lustre dead as a common solution? What is replacing it as a scalable FS in HPC applications?

    --
    There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
    1. Re:Lustre dead? by Anonymous Coward · · Score: 2, Informative

      The Lustre devs basically never played nice with the Linux Kernel devs, and Lustre never left the STAGING sub-section of the Linux Kernel because they never (in almost a decade) cleaned up the code to pass Linux Kernel code reviews.

      Lustre's most recent full release was just this last April, it's very much under heavy development, but in the HPC world having your code 'upstream' is basically a non-issue because they're used to having to do a million minor adjustments to eek out another 1% performance, since at the scale of computing they're doing that tiny improvement makes a noticeable difference. They don't run 'out of the box' Linux distros compared to a web-server or the like.

      - WolfWings, too lazy to login to /. in far too long.

    2. Re:Lustre dead? by Wrath0fb0b · · Score: 2

      The Lustre devs basically never played nice with the Linux Kernel devs, and Lustre never left the STAGING sub-section of the Linux Kernel because they never (in almost a decade) cleaned up the code to pass Linux Kernel code reviews.

      This is a major downside of the "everything in the same source tree" Linux philosophy.

      In a different universe, we could have pluggable filesystem modules that could be built against each kernel source[1] without having to get it all into the same tree. If the kernel API changed, the FS devs would have to adapt to be compatible, but at least they wouldn't have to fight with the maintainers and the maintainers wouldn't have to be bothered by their patches. It wouldn't solve a technical problem -- after all, the build system hardly gives two shits where the source comes from -- but it would solve a major human problem.

      Oh well, we can dream right?

      [1] Specifically not wanting ABI compatibility here. First, because down that road madness lies and second because it's not too much to ask that FS developers recompile when the kernel changes and adopt changing API. It also cuts off the crazy assertion that pluggable modules would stagnate the kernel code. It would, unless we just stipulate that kernel changes are absolutely allows to break any plugin and that the plugin has no right to a stable API.

      This doesn't replace in-tree, by the way. If a contributor and the maintainer agree, then by all means put it in-tree and be done with it. This is even beneficial for the contributor because it means that anyone changing the API must go and fix/shim all his code.

    3. Re:Lustre dead? by MAXOMENOS · · Score: 2
      The Register suggests that other FS's can pick up the slack. Quoting:

      When Lustre emerged in the year 2003 it had little competition for creation of large-scale filesystems. Nearly 15 years on and Red Hat offers Gluster, IBM’s Spectrum Scale (aka the GPFS General Parallel File System) and scale-out NFS can all do plenty of what made Lustre useful. HDFS has emerged, too, for big data workloads.

      Source: Linux literally loses its Lustre – HPC filesystem ditched in new kernel. The Register, 18 Jun 2018 at 01:29

      --
      Finding God in a Dog
    4. Re:Lustre dead? by afidel · · Score: 2

      He's wrong, I prefer a stable API, unstable APIs and bad vendors is why we have the mess we do with Android and why Treble is the kinda solution, by freezing on a LTSR kernel they've effectively created an artificially stable API which they backport everything else around. In an idealized world, yes I'd love to have every driver in mainline so that everything gets updated together and the kernel devs were free to monkey with things to their hearts content, in the real world I live in a stable API allows vendors to dump a driver once and for it to be usable for a decade or more (see most Windows drivers from Vista still working on 10 today).

      --
      There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
  2. Re:Causing more problems down the line by Anonymous Coward · · Score: 5, Informative

    In this case what was removed was mostly components in the staging sub-section that had been there for years (over half a decade in many cases) but hadn't completed appreciable work towards exiting the Staging sub-section. So they're finally getting the boot back out of kernel since they had never graduated to 'full kernel support' due to lack of action on their devs part. Lustre in particular basically stopped contributing to the staging branch since they felt it slowed them down too much versus working on their out-of-tree version of the code instead.

  3. Re:bug remaining? by SiChemist · · Score: 2

    I thought I had that bug, but It was fixed by switching wireless PCI cards. The ATH9K driver was the culprit on my system.

    It was pretty weird because the symptoms were the same-- Only crash at idle or near idle. As long as the CPU was under heavy load, the system ran flawlessly. I tried everything I could find online to mitigate it including the "rcu_nocbs=0-15" parameter.

    I finally noticed some ath9k messages in the logs near the time of the crash, so I bought an intel PCI wireless adapter and replaced my old one. Haven't had a crash since (that I recall). This system is up 24/7 except for kernel updates and is running Xubuntu 18.04 now. (It was on Xubuntu 16.04 at the time of the problems and my solution).

    --
    God is imaginary
  4. You need zenstates by webnut77 · · Score: 4, Informative

    How can I tell if the bug that was crashing Ryzen processors at idle has been fixed?

    In 4,15, Ryzen users had to put in a boot option

    "rcu_nocbs=0-15

    (this case for 8 cores =0-11 for a six core...)

    I had this problem too and the rcu_nocbs=0-15 didn't fix entirely the problem. You need zenstates to turn off the C6 power saving state.
    Here's the systemd unit file:

    [Unit]
    Description=Turn off power saving C6 state

    [Service]
    Type=oneshot
    StandardOutput=syslog
    ExecStart=/usr/bin/python /usr/local/src/ZenStates-Linux-master/zenstates.py --c6-disable

    [Install]
    WantedBy=basic.target

    1. Re: You need zenstates by sjames · · Score: 4, Informative

      Yes, I'm sure she'd rather see error 0×657EA6556778B4732FFED56546

      That way she can fire up her hex editor and manually patch the Windows kernel to fix the issue.

  5. "ew AMDGPU" by TeknoHog · · Score: 4, Funny

    OK, I get it that you're an Nvidia guy, but no need to rub it in my face.

    --
    Escher was the first MC and Giger invented the HR department.
  6. Re:Comment vs. "SysAdmin" scriptkiddies by sjames · · Score: 2

    No, I'm just tired of squinting at attached screenshots rather than the user with a problem just being able to cut paste 160 characters of text into an email.

    That and clearly telling the user to click on the thing that looks like a squished beetle and they keep clicking on the deformed grasshopper.

    Give us a call after you spend some time maintaining a server you have never actually touched or seen on the other side of the country, then tell us how much you like the gooey interface.

  7. Re:slashdot is a showcase for narcissists by samwichse · · Score: 2

    Whoosh

  8. Re:You I'll reply to & why? Ok... apk by sjames · · Score: 2

    You obviously haven't reached a skill level where the significant advantages of text based admin come in. Sometimes enterprise means resilliant and effective, but quite often it means brittle and catering to the lowest common denominator.

    It''s fine if users use a GUI, but the system itself should be text based so a skilled admin has a chance of actually fixing it when it's not working right. Even MS is starting to recognize that.

    As for LAN/WAN, step one when there is a problem is ditch the GUI and get a command line. If you actually understand the network and how the routers work, that will give you the tools you actually need to solve the problem. Use the GUI to check status when things are working well. If you need a picture of a thermometer to know when the temperature is too high, you'be already lost.