Announcement

Collapse
No announcement yet.

Linux 6.7 Networking Adds New Hardware Support, A ~20% Perf Boost For Single TCP Flow

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

  • Linux 6.7 Networking Adds New Hardware Support, A ~20% Perf Boost For Single TCP Flow

    Phoronix: Linux 6.7 Networking Adds New Hardware Support, A ~20% Perf Boost For Single TCP Flow

    As with each kernel cycle, the networking subsystem updates for Linux 6.7 are heavy with a wide assortment of core networking infrastructure improvements, (e)BPF features continue to be tacked on, and new wired and wireless network hardware is supported...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    phoronix smothering... the word you are looking for is smattering.

    Comment


    • #3

      - Defer regular TCP ACK while processing socket backlog, try to send a cumulative ACK at the end. Increase single TCP flow performance on a 200Gbit NIC by 20% (100Gbit -> 120Gbit).​

      It appears to me that this violates RFC requirements of sending an ACK for at least every second segment. The patch mentions a 6 fold reduction in ACKs, but does however not mention anything about RFC compliance.

      Comment


      • #4
        Originally posted by npwx View Post
        - Defer regular TCP ACK while processing socket backlog, try to send a cumulative ACK at the end. Increase single TCP flow performance on a 200Gbit NIC by 20% (100Gbit -> 120Gbit).​

        It appears to me that this violates RFC requirements of sending an ACK for at least every second segment. The patch mentions a 6 fold reduction in ACKs, but does however not mention anything about RFC compliance.
        I guess that on super high bit rate network like those, performance is more important than strict RFC compliance.

        Comment


        • #5
          the biggest part of this update is renaming a bunch of code from master/slave to conduit/user. apparently this action came from the recent netdev conference.

          Welcome back for the last day of this 10th edition of the Kernel Recipes live blog ! You can also follow the video live stream. Netconf 2023 Workshop - David Miller This year, Netconf, the invitation-only Linux conference for network kernel developers was held in parallel of the first two days


          I'm glad that they've address the blocker preventing developers from joining kernel development, and broken all documentation in the meantime.

          Comment


          • #6
            Let's erase master and slave words from history book, since both are offensive

            Comment


            • #7
              Originally posted by npwx View Post
              - Defer regular TCP ACK while processing socket backlog, try to send a cumulative ACK at the end. Increase single TCP flow performance on a 200Gbit NIC by 20% (100Gbit -> 120Gbit).​

              It appears to me that this violates RFC requirements of sending an ACK for at least every second segment. The patch mentions a 6 fold reduction in ACKs, but does however not mention anything about RFC compliance.
              no but yes in a soft way . The rule to send an ACK for at least every second segment in RFC1122 is a SHOULD and not a MUST so you are still RFC compliant if you don't. OFC there will most likely be both hw and sw out there that are built as if it was must and not should but such is life.

              Originally posted by LtdJorge View Post
              Let's erase master and slave words from history book, since both are offensive
              ​That is not what is happening so stop being silly.

              Comment


              • #8
                Originally posted by F.Ultra View Post
                ​That is not what is happening so stop being silly.
                Master and slave are words that exist, and they have a meaning. Erasing them from somewhere (where they've been used for decades already) when there are no nefarious intentions behind the choice of words is just stupid and wrong.

                Same as with blacklisting and whitelisting. Black means the total absence of light, and white means the presence of all wavelengths of light. It's a valid usage of the term, because one implies the complete absence of something in the list from the product, and the other means unequivocal presence of something in the list in the product. To think that black means bad and white means good in this use case, someone has to be either very dumb/malicious, or racist themself.

                Comment


                • #9
                  Originally posted by LtdJorge View Post
                  Master and slave are words that exist, and they have a meaning. Erasing them from somewhere (where they've been used for decades already) when there are no nefarious intentions behind the choice of words is just stupid and wrong.

                  Same as with blacklisting and whitelisting. Black means the total absence of light, and white means the presence of all wavelengths of light. It's a valid usage of the term, because one implies the complete absence of something in the list from the product, and the other means unequivocal presence of something in the list in the product. To think that black means bad and white means good in this use case, someone has to be either very dumb/malicious, or racist themself.
                  Changing the terms we use in computer lingo is not erasing words nor does "they have meaning" any relevance because so does conduit/user, and any other word for that matter. Defending the use of master/slave in computer lingo must be the dumbest hill to die on, seriously.

                  This is nothing more than enterprises not wanting to be near anything that could be deemed controversial with a ten-foot pole (like e.g GIMP), this is how the corporate world have been since inception and for good reason as well, I mean just look e.g at how some groups of people wants to cancel whole beer brands due to them believing that those brand have any other political leaning than "give us your money in exchange for this beer" after seeing some silly commercial.

                  And they are not being removed from the history books.

                  Comment


                  • #10
                    Originally posted by F.Ultra View Post

                    And they are not being removed from the history books.
                    Maybe not in the US 😞

                    Comment

                    Working...
                    X