Announcement

Collapse
No announcement yet.

Windows Server 2019 Performance Benchmarked Against Linux On An Intel Xeon Server

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

  • Windows Server 2019 Performance Benchmarked Against Linux On An Intel Xeon Server

    Phoronix: Windows Server 2019 Performance Benchmarked Against Linux On An Intel Xeon Server

    A few days back I delivered the first of our Windows Server 2019 benchmarks against Linux (as well as FreeBSD). That initial testing was done with a dual socket AMD EPYC server while in this article the tables have turned with using a dual Intel Xeon Scalable server while benchmarking Microsoft Windows Server 2019, Windows Server 2019 with its new Windows Subsystem for Linux, Windows Server 2016, and an assortment of Linux distributions including Fedora Server 29, openSUSE Leap 15, Ubuntu 18.10, CentOS 7.6, Debian 9.6, and Intel's own Clear Linux.

    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
    So the other main stream distros cannot perform any of the optimizations carried out by Clear Linux?

    Comment


    • #3
      Perhaps I missed it but the only compiler notes I see are with GCC. Were any window apps built with Microsoft compilers. If not is this really a fare comparison. A customer purchasing Xeon hardware with Windows Server will probably be more interested in Oracle or DB2 benchmarks rather then SQLite

      Comment


      • #4
        Or PostgreSQL or MySQL. I do not really see how SQLite is relevant for server performance. Maybe I'm missing some big server app that would use SQLite for data storage, but I somehow doubt that.

        Comment


        • #5
          Originally posted by DarkCloud View Post
          Perhaps I missed it but the only compiler notes I see are with GCC. Were any window apps built with Microsoft compilers. If not is this really a fare comparison. A customer purchasing Xeon hardware with Windows Server will probably be more interested in Oracle or DB2 benchmarks rather then SQLite
          On Windows, all of the official binaries are used for the apps, just as any Windows user would do. Only in the few niche cases like scimark2 and so is it built from source and in that case using GCC to be like Linux.
          Michael Larabel
          https://www.michaellarabel.com/

          Comment


          • #6
            Hi,

            what about a network performance benchmark? 10 gbit cards are cheap now, it could be usefult to see the IP switching and TCP throughput rate of the three OS

            Comment


            • #7
              Originally posted by teknoraver View Post
              Hi,

              what about a network performance benchmark? 10 gbit cards are cheap now, it could be usefult to see the IP switching and TCP throughput rate of the three OS
              Unfortunately 10 Gbit network switches still are not cheap.
              Michael Larabel
              https://www.michaellarabel.com/

              Comment


              • #8
                Originally posted by Michael View Post

                Unfortunately 10 Gbit network switches still are not cheap.
                I know, but you don't need a switch. Just four 10 gbit ports (many cards are dual slot so only two PCI devices), two on the server under test, and other two on the traffic generator.
                Traffic is sent from the generator to the first server port, routed and back to the generator, which makes the statistics.
                A good software for traffic generation like http://trex-tgn.cisco.com/ can generate dozens of millions of packets per second per core, and give very accurate results.
                That's how usually we do network benchmarks and results are quite reproducible.

                Comment


                • #9
                  Originally posted by teknoraver View Post

                  I know, but you don't need a switch. Just four 10 gbit ports (many cards are dual slot so only two PCI devices), two on the server under test, and other two on the traffic generator.
                  Traffic is sent from the generator to the first server port, routed and back to the generator, which makes the statistics.
                  A good software for traffic generation like http://trex-tgn.cisco.com/ can generate dozens of millions of packets per second per core, and give very accurate results.
                  That's how usually we do network benchmarks and results are quite reproducible.
                  ^ Yup, I have this setup at home for testing. An optical 10Gb card in each machine, with a fiber cable between them. The cards were ~$50 each on ebay, and the fiber cable was $10.

                  The cards I have are Supermicro AOC-STGN-I1S which are Intel 82599ES "X520" chipset and are plug-n-play with Linux.
                  Last edited by torsionbar28; 27 December 2018, 11:15 AM.

                  Comment


                  • #10
                    Originally posted by teknoraver View Post
                    A good software for traffic generation like http://trex-tgn.cisco.com/ can generate dozens of millions of packets per second per core, and give very accurate results.
                    That's how usually we do network benchmarks and results are quite reproducible.
                    Got any sample scripts or so? I mostly use iperf/netperf, not familiar with TRex and didn't see any very brief 'quick start guides' / 'quick examples', but if so can look at doing some benchmarks.....
                    Michael Larabel
                    https://www.michaellarabel.com/

                    Comment

                    Working...
                    X