Announcement

Collapse
No announcement yet.

The ~200 Line Linux Kernel Patch That Does Wonders

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

  • #16
    Originally posted by unimatrix View Post
    Would it be too much to ask for a PPA containing the kernel with this patch?
    And AUR script would also be nice.

    Comment


    • #17
      Originally posted by birdie View Post
      Gosh, Michael, can you use normal PCs while demonstrating improvements in kernel process management?

      Not everyone here has a six cores top Intel Core i7 CPU which costs $1000.
      I've thought about the same :-)

      My P4 is single core with two logical "cores" (HT)...

      Comment


      • #18
        $ uname -a
        Linux veyka 2.6.36-zen1-custom6-amd64 #1 ZEN SMP PREEMPT Mon Nov 15 21:12:11 GMT 2010 x86_64 GNU/Linux


        $ cat /proc/sys/kernel/sched_autogroup_enabled
        1

        running a -j4 kernel compile along with other CPU stuff, inc a game in WINE, everything is keeping going nicely, even less slowdown than before, fantastic patch!

        (It just landed in the zen kernel git last night, I only turned it on because it sounded good, glad I did!).

        Been having some more stalls at high IO load under 2.6.36. even with BFQ, which is a shame, maybe .37 will be nicer in that department.

        Comment


        • #19
          Originally posted by xeros View Post
          Ok, we see that the patch does "miracles" for heavy CPU load by compiling with many processes but...
          How about heavy disk or USB storage I/O?
          Will this patch improve desktop responsiveness when the CPU is waiting for I/O operations? If it's improving that too, that will be real benefit for most users as probably most of us use mlocate, strigi indexing or anything else indexing files in background or at least data copy between disk and usb devices.

          And what about compiz/kwin desktop effects?
          The video is showing desktop responsiveness without desktop effects.
          I personally can't comment on that one, as I've never seen more than 2-3% I/O wait (OCZ-VERTEX2).

          But desktop effects on crap graphic card (nVidia Corporation G98 [Quadro NVS 420]) under very heavy usage is as smooth as it can be. Heavy CPU usage makes no difference any more.
          Rob
          email: dagger@gentoo.org

          Comment


          • #20
            I'd *love* to try this one out but I'd need it in a PPA.

            Anyone awesome mind sharing their successful compiles?

            Comment


            • #21
              I followed this development from the first version and indeed it seems that everything is very smooth.
              The real good thing is that background tasks (system tasks) dont interfere with the user environment ('cause they are isolated in another group).

              This is achieved by the cfs scheduler that isoletes tasks belonging to the same tty.

              There is also the solution proposed by systemd or cgroups (made in user-space) but Linus stated that that would be not the solution
              1) user-space is a fragmented mess;
              2) user-space daemons are a total mess;
              3) the whole "user space is more flexible" is just a lie
              ([URL="http://lkml.org/lkml/2010/11/15/561"].

              Other than that, for what concernes IO bandwith isolation, take a look at what Mike is saying: "Next logical auto-step would be to try to subvert cfq" http://lkml.org/lkml/2010/11/15/155. Perhaps he is joking .

              Marco

              Comment


              • #22
                Why on earth was the test done with -j64... Use the amount of cores in the system. The test doesn't make sense otherwise. No one compiles kernels with -j64. Everyone uses the amount of cores plus something. Never 64.

                Comment


                • #23
                  Originally posted by RealNC View Post
                  Why on earth was the test done with -j64... Use the amount of cores in the system. The test doesn't make sense otherwise. No one compiles kernels with -j64. Everyone uses the amount of cores plus something. Never 64.
                  It was to simulate heavy load.

                  Comment


                  • #24
                    nice!

                    Anyone could share the .deb packages for the 2.6.36 kernel compiled with this patch for x64?

                    Comment


                    • #25
                      Actually, shouldn't there be a quick and easy way on Ubuntu to apt-get source the kernel, apply the patch and build the deb?

                      Comment


                      • #26
                        Originally posted by unimatrix View Post
                        Actually, shouldn't there be a quick and easy way on Ubuntu to apt-get source the kernel, apply the patch and build the deb?
                        There is. It is known as a terminal window.

                        Comment


                        • #27
                          Originally posted by Shining Arcanine View Post
                          There is. It is known as a terminal window.
                          Hahah X D

                          Not my thing playing the linux elitist, but c'mon, if people want to try something that isn't even released they should be prepared to make a minimum effort instead of crying for PPAs or what not.

                          Comment


                          • #28
                            Originally posted by yotambien View Post
                            Hahah X D

                            Not my thing playing the linux elitist, but c'mon, if people want to try something that isn't even released they should be prepared to make a minimum effort instead of crying for PPAs or what not.
                            Ubuntu has spoiled users like that now.

                            Comment


                            • #29
                              I've always avoided group scheduling. Running Gentoo, there were numerous reports of bad response in the forums beginning whenever they added group scheduling, and the basic response ever since then has been to configure it out. To be fair, I've been under the impression that group scheduling needed a userspace daemon in order to make it work right, and I've never seen anything for that other than one prototype.

                              So is this a way to make group scheduling work without the userspace daemon?

                              Wonder if there'll be a backport patch, and how far back it can go.

                              Comment


                              • #30
                                Originally posted by yotambien View Post
                                Hahah X D

                                Not my thing playing the linux elitist, but c'mon, if people want to try something that isn't even released they should be prepared to make a minimum effort instead of crying for PPAs or what not.
                                I had the same thoughts when i read those comments...

                                I am not an elitist, on the contrary i tend to help in every way i can users of a lower grade of knowledge than mine. But seriously, it is like asking to perform a brain surgery without knowing even how to do first aid...

                                If you want to play with the big boys, be willing to learn a few things first...

                                Comment

                                Working...
                                X