OK. So, even though I have to do it in a way that doesn''t seem
entirely clear,
I have managed to shape/prioritize upstream traffic. Witness:
# (sleep 7; dd if=/dev/zero bs=1M count=8 | ssh linux cat \>/dev/null &)
# ping 99.227.140.1
PING 99.227.140.1 (99.227.140.1) 56(84) bytes of data.
64 bytes from 99.227.140.1: icmp_seq=1 ttl=255 time=8.62 ms
64 bytes from 99.227.140.1: icmp_seq=2 ttl=255 time=6.91 ms
64 bytes from 99.227.140.1: icmp_seq=3 ttl=255 time=6.70 ms
64 bytes from 99.227.140.1: icmp_seq=4 ttl=255 time=12.7 ms
64 bytes from 99.227.140.1: icmp_seq=5 ttl=255 time=6.65 ms
64 bytes from 99.227.140.1: icmp_seq=6 ttl=255 time=5.83 ms
64 bytes from 99.227.140.1: icmp_seq=7 ttl=255 time=9.73 ms
64 bytes from 99.227.140.1: icmp_seq=8 ttl=255 time=110 ms
64 bytes from 99.227.140.1: icmp_seq=9 ttl=255 time=52.5 ms
64 bytes from 99.227.140.1: icmp_seq=10 ttl=255 time=33.3 ms
64 bytes from 99.227.140.1: icmp_seq=11 ttl=255 time=34.4 ms
64 bytes from 99.227.140.1: icmp_seq=12 ttl=255 time=13.6 ms
64 bytes from 99.227.140.1: icmp_seq=13 ttl=255 time=61.0 ms
64 bytes from 99.227.140.1: icmp_seq=14 ttl=255 time=41.0 ms
64 bytes from 99.227.140.1: icmp_seq=15 ttl=255 time=17.3 ms
64 bytes from 99.227.140.1: icmp_seq=16 ttl=255 time=20.3 ms
64 bytes from 99.227.140.1: icmp_seq=17 ttl=255 time=25.4 ms
So as you can see when pining the next-hop router (99.227.140.1), at the 7-8s
mark there is a tiny blip upward, and even while that is still reasonable, ping
times come back even better. The shaping is working wonderfully.
For reference, the host "linux" to which I''m pushing traffic
in the benchmark,
is actually at the other end of an openvpn tunnel from the machine I am running
the above on, so the upstream bandwidth is actually saturated with openvpn udp
packets, not raw ssh packets and so the shaping on the external interface (eth0)
is dealing with the openvpn packets and the icmp packets and doing a wonderful
job.
What doesn''t seem to be working is shaping inside the tunnel. Witness:
# (sleep 7; dd if=/dev/zero bs=1M count=8 | ssh linux cat \>/dev/null &)
# ping linux
PING 10.75.22.3 (10.75.22.3) 56(84) bytes of data.
64 bytes from 10.75.22.3: icmp_seq=1 ttl=63 time=34.0 ms
64 bytes from 10.75.22.3: icmp_seq=2 ttl=63 time=31.8 ms
64 bytes from 10.75.22.3: icmp_seq=3 ttl=63 time=32.4 ms
64 bytes from 10.75.22.3: icmp_seq=4 ttl=63 time=35.4 ms
64 bytes from 10.75.22.3: icmp_seq=5 ttl=63 time=30.9 ms
64 bytes from 10.75.22.3: icmp_seq=6 ttl=63 time=32.3 ms
64 bytes from 10.75.22.3: icmp_seq=7 ttl=63 time=572 ms
64 bytes from 10.75.22.3: icmp_seq=8 ttl=63 time=763 ms
64 bytes from 10.75.22.3: icmp_seq=9 ttl=63 time=150 ms
64 bytes from 10.75.22.3: icmp_seq=10 ttl=63 time=181 ms
64 bytes from 10.75.22.3: icmp_seq=11 ttl=63 time=367 ms
64 bytes from 10.75.22.3: icmp_seq=12 ttl=63 time=586 ms
64 bytes from 10.75.22.3: icmp_seq=13 ttl=63 time=706 ms
64 bytes from 10.75.22.3: icmp_seq=14 ttl=63 time=767 ms
64 bytes from 10.75.22.3: icmp_seq=15 ttl=63 time=775 ms
64 bytes from 10.75.22.3: icmp_seq=16 ttl=63 time=776 ms
64 bytes from 10.75.22.3: icmp_seq=17 ttl=63 time=877 ms
64 bytes from 10.75.22.3: icmp_seq=18 ttl=63 time=1435 ms
64 bytes from 10.75.22.3: icmp_seq=19 ttl=63 time=718 ms
64 bytes from 10.75.22.3: icmp_seq=20 ttl=63 time=891 ms
So same data generation, inside the tunnel and a ping also inside the tunnel.
Now the traffic shaping on eth0 is dealing with only the openvpn udp packets and
I need traffic shaping/prioritization of the traffic in the tunnel, on tun0.
http://www.shorewall.net/simple_traffic_shaping.html suggests being able to name
a VPN interface in the tcinterfaces to effect shaping on that interface as such:
#INTERFACE TYPE IN-BANDWIDTH
tun0 Internal
Which I have done:
# cat tcinterfaces
#INTERFACE TYPE IN-BANDWIDTH
eth0 External 6mbit 500kbit
tun0 Internal
But that doesn''t seem to give me those nice results as
shaping/prioritization on
eth0. What am I missing?
For the record, the openvpn traffic is the only traffic on the external
interface (eth0) so this is not a case of the openvpn traffic contending with
other traffic.
Any ideas?
Cheers,
b.
------------------------------------------------------------------------------
RSA(R) Conference 2012
Save $700 by Nov 18
Register now
http://p.sf.net/sfu/rsa-sfdev2dev1