Why does the bare metal's bandwidth decrease after installing Docker on Windows Server 2016? RRS feed

  • Question

  • I want to measure the Windows Container performance and met a strange problem.

    I have two servers installing Windows Server 2016(Standard Edition) and use NTttcp measuring the bindwidth of two servers. If the server don't install the Docker(17.03.01-ee-3), the bandwidth of two servers is 17-18Gbps. However, the bandwidth of two servers is 6Gbps when the server has installed Docker. I have no idea why the Docker will affect the bandwidth of the bare metal. 

    Moreover, if I make the server(installed docker) as sender(the receiver doesn't install docker), the bandwidth will decrease. But if i make the server(installed docker) as receiver(the sender doesn't install docker), the bandwidth isn't affected. I am so confused...

    Does anybody meet the same question? Why does the bare metal's bandwidth decrease after installing Docker on Windows Server 2016?

    Wednesday, May 3, 2017 5:16 AM

All replies

  • Assuming you are the same Flying Dove as posted in the Docker forums?  The Docker folks pinged us on it this morning.

    Thanks for posting here as well.  Kallie and the networking team are following up on our side.

    This is, as you say, a weird one.  

    I suppose there is also a small chance this has to do with the way NTttcp measures bandwidth, and not actual networking performance.  Have you tried just doing things with the different installation setups?  Is there a noticeable performance hit?

    Either I or someone else on the team will come back with more ideas if we can reproduce the issue.  

    Of course, please feel free to bump this or add new info and I'll respond.



    Wednesday, May 3, 2017 7:36 PM
  • Thank you first! 

    The problem really annoys me for a long time. If it really like what you said, i don't konw how i should measure the 'actual network performance' if i don't use NTttcp? I also tried to use the tool called 'iperf' to measure it but the performance are even worse.

    As the different installation setups, to be honest i am not very clear about it.I believe that i just use one setup to install windows and docker(I follow the guide of this website:

    Expect your reply, thanks!

    Thursday, May 4, 2017 1:37 PM
  • That guide is our doc, great place to start.

    By "actual network performance" I mean that you are setting up these servers to do something.  I'm guessing your ultimate goal is not just to run performance tools.  

    Rather I'm guessing you want to run some workload.

    Have you tried running that workload with these servers?  Does this reduction in bandwidth cause issues?

    Kallie and the other networking folks are working on this one.  They are very busy with prep for Build 2017 so please be patient.

    Friday, May 5, 2017 9:05 PM
  • I haven't used these server running some workloads yet, and I will try later.

    Thank you for your reply! 

    Monday, May 8, 2017 5:31 AM
  • Great!  Let me know how it goes :)
    Tuesday, May 9, 2017 10:18 PM