Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror
×
GNU is Not Unix Graphics Open Source Operating Systems Power Software Upgrades Linux

Linux 4.6 Brings NVIDIA GTX 900 Support, OrangeFS, Better Power Management (phoronix.com) 129

An anonymous reader writes: The Linux 4.6-rc1 kernel has been released. New to the Linux 4.6 kernel are a significant number of new features including NVIDIA GeForce GTX 900 open-source 3D support when using the closed-source firmware files, Dell XPS 13 Skylake laptop support, a fix for laptops that were limiting their own performance due to incorrectly thinking they were overheating, AHCI runtime power management support, Intel graphics power management features enabled by default, a new file-system (OrangeFS), and a range of other improvements.
This discussion has been archived. No new comments can be posted.

Linux 4.6 Brings NVIDIA GTX 900 Support, OrangeFS, Better Power Management

Comments Filter:
  • OrangeFS (Score:5, Funny)

    by Anonymous Coward on Saturday March 26, 2016 @09:37PM (#51784997)

    Developers provided performance charts against AppleFS in the release notes, but they found it wasn't comparing the same thing.

    • I looked over the homepage for OrangeFS and still can't figure out the point.

      • Re:OrangeFS (Score:5, Informative)

        by Anonymous Coward on Saturday March 26, 2016 @10:00PM (#51785087)

        From a glance at their website, I think the keyword is MPI-IO. Picture a weather simulation consisting of 10,000 processes that work together in parallel, where all of them perform semirandom input/output using the same files. That's gonna be difficult to do right with conventional filesystems...

        • Your post was 1000 times more informative than their site. Thank you.

        • Re: (Score:2, Funny)

          That seems like a crazy waste of energy. If you want to predict the weather approximately as accurately as the local news, just roll a d20 and look the results up in your DM manual.

        • When people here filesystem they think of a disk file system.

          Maybe calling it Orange Clustering System might be a more descriptive name?

        • by Khyber ( 864651 )

          "Picture a weather simulation consisting of 10,000 processes that work together in parallel, where all of them perform semirandom input/output using the same files. That's gonna be difficult to do right with conventional filesystems...'

          But absolutely trivial to do on a RAMDisk. This seems like a solution in search of a problem, however said problem was solved long ago with far superior capability.

          • Comment removed based on user account deletion
            • by Anonymous Coward

              Do explain how to share that RAMdisk between tens of thousands of processes running on multiple different systems, possibly handling petabytes of data and needing redundancy, too? Oh, right, RAMdisks can't handle such needs.

              If the same file is accessed by multiple nodes, this implies no data moves. That is, each part of the file has a specific meaning. When you're building a cluster for such a purpose, there likely is no need for multiple files. Or, for performance reason, noone sane would put multiple files on the same disk. Speed will be needed, and RAM can easily be build as big as today's SSDs, even server ones. What's needed is simple direct access to the same block device, with no file system getting in the way at all.

          • And what filesystem do you put on that ramdisk that allows access from 10,000 different machines, simultaneously? And how does that single machine connect to the network? Do you have access top a machine with 1PB of RAM? We have a few 1.5TB machines, but our filesystem is more like 1TB. Basically, this has nothing to do with ram disks, other than that the servers will have huge gobs of ram cache.
            • by Khyber ( 864651 )

              When you make a RAMDisk for multiple machine access, you give it no FS and just give it raw data with pointers to that data when requested.

              It's exactly how my 2D Second Life clone works. Very efficient, very fast, all world changes are practically instantaneous. One raw binary resource file, multiple nodes access it all from RAM, every access forces a refresh across every accessing node. You only need a proper backplane for node interconnectivity.

      • Seriously, there seems to be an obsession in parts of the Linux community to make yet another thing that already does what a bunch of existing software does. I'm not sure why but you see a lot of it. You'll get a distro that'll have 8 different media players, none of them worth a damn, rather than just having one good one but hey, you have options!

      • Re:OrangeFS (Score:4, Informative)

        by flyingfsck ( 986395 ) on Sunday March 27, 2016 @09:11AM (#51786587)
        OrangeFS is a derivative of the parallel FS originally used for Beuwolf clusters.
  • Is that a good thing?

    • by Anonymous Coward

      In general: Maybe. It's customary to have a C version and have assembly versions "override" the C as and when needed. If all you had was the assembly version then a C version would likely be nice to have around. If it's a piece of code that only makes sense on one architecture anyway and the replacement runs half the speed and double the code size, that effort might well have been better spent elsewhere, even if the code wasn't performance critical.

      In this case... well, it appears to be a maintenance issue,

Keep up the good work! But please don't ask me to help.

Working...