-2

According to the following paper...

COMPARING NETWORK PERFORMANCE: RED HAT ENTERPRISE LINUX 6 VS. MICROSOFT WINDOWS SERVER 2012

Src: http://www.principledtechnologies.com/Red%20Hat/RHEL6_network_0613.pdf

... Windows Server network performance could be as much as 3 times slower than Linux.

Anyone knows another benchmark where Windows Server stack is on par with Linux?

Maybe the situation with slow networking got improved in latest Windows Server 2019?

Dmitry Sychov
  • 119
  • 4
  • Why do you care about one synthetic networking benchmark from two major OS versions ago? 1) Latest OSes will be different 2) Your workload will be different 3) There are more important factors than microsecond differences in TCP round trips – John Mahowald Jun 16 '20 at 03:41
  • I failed to find any Windows Server benchmarks - how good it scales? how good it saturates top NIC cards? That's the "best" info comparing network stacks I was able to get. – Dmitry Sychov Jun 16 '20 at 16:30
  • Also: https://superuser.com/questions/922729/is-it-possible-to-process-millions-of-datagrams-per-second-with-windows – Dmitry Sychov Jun 16 '20 at 16:32
  • I need some data comparing Windows & Linux networks stacks - is Windows as bad as it sounds? 4Mpps overall is not that much if they were using all cores since Linux is capable of pushing 1..2 Mpps per core with excellent scaling. I'am afraid Windows Server could lag behind as much as 10x times in overall scaling(mpps) over multicore cfgs with hundreds of cpu cores(think Epyc, etc). And no one to date tried to put Windows Server to honest test... – Dmitry Sychov Jun 16 '20 at 16:34
  • 2
    But you can test it yourself! – Michael Hampton Jun 16 '20 at 20:01
  • @Michael Hampton have a pair of epyc servers with 100gbe nics? – Dmitry Sychov Jun 16 '20 at 23:44
  • 1
    @DmitrySychov Is that what you intend to run? Then go buy them. Otherwise use the hardware you are really going to use in production. – Michael Hampton Jun 17 '20 at 00:33
  • When you are buying sport motorbike from the dealer are you doing this based on certain model characteristics from manufacturer or just to test run? – Dmitry Sychov Jun 17 '20 at 10:54
  • 1
    Test drives are a thing. In this case, some proof of concept equipment from the sales engineers to convince you. Perhaps some 25 Gb Ethernet (2x faster than that benchmark) but you haven't explained what your needs are. – John Mahowald Jun 18 '20 at 02:51

1 Answers1

1

Years-old micro benchmarks are not how to select an operating system.

Your workload will not be like that benchmark.

Both platforms have had two major releases since then. Network stacks will perform differently in the latest releases.

Other factors in the choice can be just as important. The software running on it, experience managing it, and your support relationship.

If you need to push Windows to multi-million pps, try it and see. Involve performance engineers from your hardware vendor, the NIC manufacturer, and Microsoft. Should it be necessary, pick a different OS that meets your requirements.

John Mahowald
  • 30,009
  • 1
  • 17
  • 32
  • MS should publish the data on their network stack. Or its a pure speculation thinking they've improved in the last decade. We are trying to contact NIC manufacturers already. Switching to diff. OS(diff IDE, diff API) for years just because MS did not bother publish benchmarks comparing its network stack to the latest linux implementation? that's totally insane to say the least. – Dmitry Sychov Jun 16 '20 at 23:43