Daniel Lezcano <dlezcano@fr.ibm.com> writes:

> Hi Herbert,
>
> I played with the L2 namespace patchset from Eric Biederman, I did some
> benchmarking with netperf:
>
> With 2 hosts, Intel EM64T bipro HT / 2,4 GHz , 4Go ram and GB network.
> Host A is running the netserver on a RH4 kernel 2.6.9-42
> Host B is running the netperf client inside and outside the container with the
> command:
> netperf -H HostA -c -l 20 -n 2 -p 12865
>
> Results are:
> inside the container:
>    Throughput : 940.39  Mbit/s CPU usage : 15.80 %
>
> outside the container:
>    Throughput : 941.34 Mbits/s CPU usage :  5.80 %

Could you clarify a little bit what you mean by inside the container
and outside the container?   In particular was the outside the container case
in a kernel that had the network namespace compiled in?

Could you also clarify how you have setup networking inside the container
and going to the outside world?

> I did the test again with 50 containers. I created them one by one having one
> running netperf and the other being idle.
> Each time I created a container, I rerun netperf. To be more explicit, I created
> 1 container, run netperf inside it and blocked it on a fifo reading, I created a
> second container, run netperf inside it and blocked it, and son on ... to 50
> containers. The benchmarking result are the same as running one container, so I
> guess it scales well.
>
> There are a lot of scenarii to do for benchmarking, for example, running netperf
> in each container in the same time and look how it behaves.
> I am profiling the kernel to look where the cpu overhead is.

Thanks.  This is simple enough that at least part of this should be easy
to reproduce.  Once things settle down a little I'm interested in tracking
the cpu usage if someone else hasn't done it already.

Eric
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers