1. openstack version: newton(same with master)
2. neutron --version 11.0.1
3. ovs-vsctl (Open vSwitch) 2.7.2
4. iperf3
The ovs config is as follow:
[root@cdpm04 common]# ovs-vsctl list qos
_uuid : b90a11e9-ea70-4908-8cb2-bab21e117312
external_ids : {id="qvocd3421e0-43"}
other_config : {max-rate="3000000"}
queues : {0=22ba7a96-328c-4876-b7ac-b35d77ce3a09}
type : linux-htb
[root@cdpm04 common]# ovs-vsctl list queue
_uuid : 22ba7a96-328c-4876-b7ac-b35d77ce3a09
dscp : []
external_ids : {id="qvocd3421e0-43", queue_type="0"}
other_config : {burst="300000", max-rate="3000000"}
The burst does not take effect.Even I change the value of burst, the result does not change.
The result is as follows:
[ ID] Interval Transfer Bandwidth Retr
[ 5] 0.00-11.27 sec 4.50 MBytes 3.35 Mbits/sec 0 sender
[ 5] 0.00-11.27 sec 3.87 MBytes 2.88 Mbits/sec receiver
I think this problem is related to [1][2].
Maybe we do not have to change like that.After I revert the code, the result is accurate and the burst take effect.
[1]https://bugs.launchpad.net/neutron/+bug/1730605
[2]https://review.openstack.org/519573
I was checking it on my devstack and I have couple of questions to You about it.
1. Is Your problem with ingress or egress traffic (from instance PoV)? In bug title You said egress but qos and queue is set in ovs for ingress bw limit only.
2. For tcp traffic burst should be set for about 80% of bandwidth rate. So in Your example it should be about 2400000 instead of 300000.
3. Result which You paste here looks quite good for me. What results You are expecting exactly?
4. Can You give me more details how to reproduce this issue exactly? So output of how exactly Your bw limit rule was created in Neutron, between what nodes You are doing test (two instances or instance and some outside server) and also how exactly You run iperf3 client and server on those test nodes.