Re: Investigating network performance when eBPF is in use

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



I found the problem. When I added the tun0 interface to the bridge,
performance increased to more than 10000 TCP_CRR round trips per
second.

I needed to create tun0 in the root namespace, add it to the bridge,
and then move it to the namespace where it would be used.
ip link set tun0 master br0
ip link set dev tun0 netns ns2

Chris Ward, IBM

On Mon, 28 Nov 2022 at 16:47, Chris Ward <tjcw01@xxxxxxxxx> wrote:
>
> I have a test case which shows unexpectedly poor performance for one
> scenario (TCP_CRR netperf test between 2 namespaces on 1 machine).
> This test case delivers under 1 round-trip per second, where it should
> be delivering more than 10000 round-trips per second.
> I have tried disabling Nagle, but that does not increase the observed rate.
>
> Are there any tools I should be using to investigate what is happening
> ? Or does anyone on this lisk know what is going on and how to fix it
> ?
>
> My test case is available here
> https://github.com/tjcw/bpf-examples/tree/tjcw-integration-1.2/AF_XDP-filter/netperf-namespace
> if anyone wants to try repeating my results.
>
> Chris Ward, IBM



[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux