Hi, I am facing problem with bandwidth slicing when I use TCP traffic. I use OVS v 1.10 as OpenFlow switch, and today I connected three hosts on it in order to test OVS qos mechanisms. This is topology I have used:
iperf client 1 ---- OVS---- iperf server
I
iperf client 2
On the port towards iperf server I defined QoS policy like this:
sudo ovs-vsctl -- set port eth4 qos=@newqos -- --id=@newqos create qos
type=linux-htb other-config:max-rate=10000000 queues=1=@q1,2=@q2 --
--id=@q1 create queue other-config:min-rate=7000000
other-config:max-rate=10000000 -- --id=@q2 create queue
other-config:min-rate=3000000 other-config:max-rate=10000000
When I generate UDP traffic with iperf clients everything works fine.
However, when I generate TCP traffic average measured bandwidth is ok
,if I generate traffic for some longer period (30 s or more) , but the
way how this is achieved is disaster. I adjusted iperf to report me
state every second and for few consecutive seconds reported value for
one of the clients was 0 Mb/s, then after some time 10 (I wanted
3Mbit/s) and so on... I expected that with TCP would be bandwidth
fluctuations but not like these.
When I performed similar test with Stanford Reference OF switch I
didn't have this problem. However, OVS has more capabilities and I
need it for application that I am developing.
Can someone tell me can I somehow fix this?
I am sending you in an attachment iperf results for client1 and client2.
Thanks in advance!
Slavica
client2
Description: Binary data
client1
Description: Binary data
_______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
