Hi,
I was testing the impact over the flow setup latency when the channel between the OVS and the controller is congested. I have defined three time intervals: the time that a PKT_IN takes to arrive at the controller (B1), the time to process the PKT_IN and send a FLOW_MOD (B2), and the time that the FLOW_MOD takes to arrive at the OVS (B3). When the channel is congested, a new iperf connection is started to measure the latency. Whereas the congestion is greater, the value of B1 is greater too, but B2 must be the same, i.e. the time to process a PKT_IN must not be modified by the congestion. Strangely, Pyretic increments the time to process a PKT_IN when the congestion is greater. Apparently there is a dependency between the channel's congestion and the PKT_IN's processing time.
I'm not sure if this is a expected behaviour, because this same test using another controller shows a constant values of B2, which is a correct expected behavior.
Thanks in advance for your help.
Ricardo.
Hi, I was testing the impact over the flow setup latency when the channel between the OVS and the controller is congested. I have defined three time intervals: the time that a PKT_IN takes to arrive at the controller (B1), the time to process the PKT_IN and send a FLOW_MOD (B2), and the time that the FLOW_MOD takes to arrive at the OVS (B3). When the channel is congested, a new iperf connection is started to measure the latency. Whereas the congestion is greater, the value of B1 is greater too, but B2 must be the same, i.e. the time to process a PKT_IN must not be modified by the congestion. Strangely, Pyretic increments the time to process a PKT_IN when the congestion is greater. Apparently there is a dependency between the channel's congestion and the PKT_IN's processing time. I'm not sure if this is a expected behaviour, because this same test using another controller shows a constant values of B2, which is a correct expected behavior. Thanks in advance for your help. Ricardo.