[lxc-devel] limit the netwok traffic of container from the host

[Serge Hallyn]

Quoting lsmushroom (lsmushroom at 126.com):
> Sorry,  I really get confused. Can you show me your testing procedure ?
> Below is my testing result , the veth pair name is veth-vps1:
> 
> 
> 1) Add qdisc on veth-vps1
> tc qdisc add dev veth-vps1 root tbf rate 0.05mbit burst 5kb latency 70ms peakrate 1mbit minburst 1540
> 
> 
> 2) Send packet from the container, run iperf server on the host end, and iperf client inside the container: 
> [root@ iperf_2.0.2-4_amd64]# ./iperf -s -p 10086 -i 1
> ------------------------------------------------------------
> Server listening on TCP port 10086
> TCP window size: 85.3 KByte (default)
> ------------------------------------------------------------
> [  4] local 172.16.10.5 port 10086 connected with 172.16.10.125 port 38580
> [  4]  0.0- 1.0 sec  1.76 MBytes  14.7 Mbits/sec
> [  4]  1.0- 2.0 sec  1.03 MBytes  8.63 Mbits/sec
> [  4]  2.0- 3.0 sec  1.18 MBytes  9.86 Mbits/sec
> [  4]  3.0- 4.0 sec  1.17 MBytes  9.81 Mbits/sec
> [  4]  4.0- 5.0 sec  1.26 MBytes  10.6 Mbits/sec
> [  4]  5.0- 6.0 sec  1.13 MBytes  9.45 Mbits/sec
> [  4]  6.0- 7.0 sec  1.16 MBytes  9.71 Mbits/sec
> [  4]  7.0- 8.0 sec  1.20 MBytes  10.0 Mbits/sec
> [  4]  8.0- 9.0 sec  1.20 MBytes  10.1 Mbits/sec
> [  4]  0.0- 9.6 sec  11.7 MBytes  10.3 Mbits/sec
> 
> 
> 3) Send packet to the container, run iperf client on the host end, and iperf server inside the container: 
> [root@ iperf_2.0.2-4_amd64]# ./iperf -s -p 10086 -i 1
> ------------------------------------------------------------
> Server listening on TCP port 10086
> TCP window size: 85.3 KByte (default)
> ------------------------------------------------------------
> [  4] local 172.16.10.125 port 10086 connected with 172.16.10.5 port 34648
> [  4]  0.0- 1.0 sec  5.66 KBytes  46.3 Kbits/sec
> [  4]  1.0- 2.0 sec  7.07 KBytes  57.9 Kbits/sec

Thanks.

So yeah, this clearly answer's my question which was:

> >Are you saying it will limit only traffic to, but not from, the
> >container?

(I don't doubt your test, won't reproduce it right now :)

Now what you've proposed would solve this in a more flexible way,
allowing traffic both into and out of the container to be slowed down
independently.  But on the other hand, especially given the scant amount
of information on tc out there, I think many people would end up
misconfiguring.

Do you think it would be an ok idea to have 'qdisc add dev <some-veth>'
add the restriction on both of the tunnel end-points?

(The answer may well be no, but I do prefer the simpler configuration)

-serge