As I mentioned, I have 2 servers connected via 1Gig cable. I have installed OVS with DPDK on server one. I have a switch OVS_BR_LEAF_1 on it which runs in user-space. I tried doing QoS traffic shaping as done in the link you mentioned.
I have following flow entries in my switch #vs-ofctl add-flow OVS_BR_LEAF_1 priority=6000,in_port=LOCAL,actions=set_queue:1,normal #ovs-ofctl add-flow OVS_BR_LEAF_1 priority=6000,in_port=2,actions=normal I have connected eth3 to ovs on port 2. I want traffic going from server one to server two to observer 20Mbps traffic rate. To deconfigure any old QoS record from eth3 and then delete the QoS record, I did: #ovs-vsctl clear Port eth3 qos Then, I added following configuration: #ovs-vsctl -- set port eth3 qos=@newqos -- --id=@newqos create qos type=linux-htb other-config:max-rate=500000000 queues:1=@q1 -- --id=@q1 create queue other-config:max-rate=20000000 Then I checked rate via iperf and got around 478Mbps, which is closer to the max rate of the default queue i.e. 500000000. I have checked it multiple times with different values of default queue's max-rate, I just get whatever max-rate of default queue I specify, the queue q1 never works. Please help. On Thu, Dec 13, 2018 at 5:47 AM Justin Pettit <[email protected]> wrote: > > > On Dec 12, 2018, at 12:32 AM, Ramzah Rehman <[email protected]> > wrote: > > > > I have tested meter action in OpenFLow. I have drawn following > conclusions: > > > > • TCP > > • Conclusion: meter works fine for low rate-limiting > values (till ~100Mbps). However, for higher values, the expected bandwidth > is relatively very low. > > • UDP: > > • Conclusion: meter action not supported for UDP > > > > Please let me know Why I am seeing this behavior with TCP traffic? Is it > expected? If so, How can I get expected bandwidth equal to the rate I > specify in OpenFlow meter? > > TCP throughput has a tendency to act erratically to packets dropping, so > it can be hard to exactly tune throughput. You might try configuring > different burst sizes at different rates to see if that helps. However, if > you need to rate-limit TCP, you're probably better off using the QoS > configuration with traffic shaping, since that will delay packets instead > of dropping them. Some of this is discussed in the FAQ: > > http://docs.openvswitch.org/en/latest/faq/qos/ > > I would recommend using HTB for less than 100Mbps and HFSC for less than > 1Gbps. Anything above that will likely have poor performance due to the > overhead of performing QoS in software. > > > Also, does meter action work for UDP traffic? > > Yes, I would expect it to work. It should perform more predictably than > TCP. > > How did you configure the flows? Are you using the kernel or userspace > datapath? If you're using the kernel, what kernel version are you using? > There are a few kernel versions that have issues due to a bug in the > upstream kernel. > > --Justin > > >
_______________________________________________ discuss mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
