Announcement

Collapse
No announcement yet.

Intel Developer Proposes "Kernel NET Policy" For Better Linux Network Performance

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

  • Intel Developer Proposes "Kernel NET Policy" For Better Linux Network Performance

    Phoronix: Intel Developer Proposes "Kernel NET Policy" For Better Linux Network Performance

    Intel developer Kan Liang has published a set of 30 patches amount to more than two thousand lines of new kernel for implementing what he calls the Kernel NET Policy...

    http://www.phoronix.com/scan.php?pag...nel-NET-Policy

  • #2
    I would be interested in how to get optimal performance for gaming.
    This is where low latency is of uttermost most importance, but bandwidth is not.

    Comment


    • #3
      Sounds interesting. Network tuning has always been less-than-fun.

      Originally posted by uid313 View Post
      I would be interested in how to get optimal performance for gaming.
      This is where low latency is of uttermost most importance, but bandwidth is not.
      the latency that matters in gaming is the router-game-servers/peers, not the gaming-PC-router one.

      Comment


      • #4
        I would love to see the kernel improved to a point where it can go up against Cisco IOS. It's one area where Linux hasn't gotten a foothold yet...at least as far as I know.

        Comment


        • #5
          This looks like a very nice framework for actually giving users more control to tweak the network interfaces the way they want.
          I hope it becomes more widespread for other high-speed interfaces too.

          Comment


          • #6
            Originally posted by uid313 View Post
            I would be interested in how to get optimal performance for gaming.
            This is where low latency is of uttermost most importance, but bandwidth is not.
            You want to be looking into fq_codel / Cake then.

            Comment


            • #7
              Originally posted by starshipeleven View Post
              Sounds interesting. Network tuning has always been less-than-fun.

              the latency that matters in gaming is the router-game-servers/peers, not the gaming-PC-router one.
              I heard World of Warcraft players and other players use something called Leatrix Latency Fix on OS X and Windows.

              Comment


              • #8
                Originally posted by starshipeleven View Post
                Sounds interesting. Network tuning has always been less-than-fun.

                the latency that matters in gaming is the router-game-servers/peers, not the gaming-PC-router one.
                Truly spoken like someone who hasn't done much gaming. QoS programs and high quality NICs were all the rage in the 2000's just because of that reason. I don't know how linux's network latency compares to windows, but the window's default settings and onboard NIC performance as far as latency goes were quite crap, and probably still are. 10-20ms difference is quite large in FPS games.

                Comment


                • #9
                  Originally posted by SaucyJack View Post

                  Truly spoken like someone who hasn't done much gaming. QoS programs and high quality NICs were all the rage in the 2000's just because of that reason. I don't know how linux's network latency compares to windows, but the window's default settings and onboard NIC performance as far as latency goes were quite crap, and probably still are. 10-20ms difference is quite large in FPS games.
                  Throughput is a lot better, latency is the same.

                  On my Gigabit Fiber, I can run a speedtest on any of my Linux PCs and it will hit 800+ Mb/sec both up and down to the Internet within two seconds.

                  Run the same speedtest on my Windows Vista PC and it will start at 100 Mb/sec and slowly increase to 800+Mb / sec over 10 seconds. It does the same thing with file transfers over the LAN. Boot it into Linux and it will hit 800+ Mb/sec both up and down to the Internet within two seconds.

                  Ran the test again on my other Linux PC and got the same results. Even the dual-core 2.4 Ghz linux PC running an AM2 socket will speed up to 800+Mb/sec within two seconds. I don't get why Windows is so slow to speed up.

                  First hop to the Internet, and out of google's network, which is what really matters the most, is around 18 ms. So that will always be my minimum ping for multiplayer. It's the same on Windows and linux.

                  Nothing fancy about the linux kernel I'm running. 3.16 on Debian stable. nVidia chipset on one linux box, and AMD chipsets on the other linux box and laptop. They behave the same. Doesn't seem hardware has a lot to do with it.
                  Last edited by Sidicas; 07-18-2016, 08:00 PM.

                  Comment


                  • #10
                    Originally posted by Sidicas View Post

                    Throughput is a lot better, latency is the same.

                    On my Gigabit Fiber, I can run a speedtest on any of my Linux PCs and it will hit 800+ Mb/sec both up and down to the Internet within two seconds.

                    Run the same speedtest on my Windows Vista PC and it will start at 100 Mb/sec and slowly increase to 800+Mb / sec over 10 seconds. It does the same thing with file transfers over the LAN. Boot it into Linux and it will hit 800+ Mb/sec both up and down to the Internet within two seconds.

                    Ran the test again on my other Linux PC and got the same results. Even the dual-core 2.4 Ghz linux PC running an AM2 socket will speed up to 800+Mb/sec within two seconds. I don't get why Windows is so slow to speed up.

                    First hop to the Internet, and out of google's network, which is what really matters the most, is around 18 ms. So that will always be my minimum ping for multiplayer. It's the same on Windows and linux.

                    Nothing fancy about the linux kernel I'm running. 3.16 on Debian stable. nVidia chipset on one linux box, and AMD chipsets on the other linux box and laptop. They behave the same. Doesn't seem hardware has a lot to do with it.
                    Unfortunately you can't really test like that. You need to do it under light load conditions like a game rather than no load. Easiest way is to connect to a reliable server and monitor the connection.

                    Comment

                    Working...
                    X