Announcement

Collapse
No announcement yet.

Another Sizable Performance Optimization To Benefit Network Code With Linux 5.17

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

  • Another Sizable Performance Optimization To Benefit Network Code With Linux 5.17

    Phoronix: Another Sizable Performance Optimization To Benefit Network Code With Linux 5.17

    Last week I wrote about a big TCP performance optimization having been queued up into net-next for Linux 5.17. That optimization can yield significant TCP throughput improvements especially with today's high-end 100Gb+ network hardware. There is now another separate juicy optimization to benefit the Linux network performance in the next kernel cycle...

    https://www.phoronix.com/scan.php?pa...r-csum_partial

  • #2
    I for one welcome our new exciting networking perf!
    hopefully we can start to get closer to top speed

    Comment


    • #3
      @Michael: you should add a note that this benefit is only for x86.
      Last edited by mlau; 23 November 2021, 07:22 AM.

      Comment


      • #4
        I hope it won't regress the 100Mbit-1Gbit connections.

        Comment


        • #5
          Originally posted by mlau View Post
          @Michael: you should add a note that this benefit is only for x86.
          It's mentioned in the article it's x86/x86_64.
          Michael Larabel
          http://www.michaellarabel.com/

          Comment


          • #6
            Would be interesting to see a performance comparison of FreeBSD vs Linux network stacks.

            Comment


            • #7
              Originally posted by Yalok View Post
              Would be interesting to see a performance comparison of FreeBSD vs Linux network stacks.
              I went to see how this do against windows 10/11

              Comment


              • #8
                Originally posted by cewbdex View Post
                I hope it won't regress the 100Mbit-1Gbit connections.
                Why would it? This patch improves the TCP/IP stack across the board. Now the practical difference will be quite small since calculating the csum is just a tiny portion of the code and there are some drivers that offload it to the NIC.

                Comment


                • #9
                  Originally posted by cewbdex View Post
                  I hope it won't regress the 100Mbit-1Gbit connections.
                  From the looks of it, shouldn't it just reduce CPU overhead for all NICs? Maybe this would even improve throughput for very weak processors.

                  Comment


                  • #10
                    There are a big elephant in the room..
                    amd64 is little endian,network is big endian.. and so all bits need to be swapped in and out, for every package, in high speed networks that takes its tall..we are already at 400Gbps, there will be a time were you will need "30 cores just for that"..

                    Comment

                    Working...
                    X