Bugzilla – Bug 2394
FlowMonitor gets confused with traffic control requeuing
Last modified: 2016-08-28 10:46:37 UTC
FlowMonitor records the drops from the netdevice and do not consider the requeuing done by the traffic control layer. I think the problem is related with the issue mentioned here http://mailman.isi.edu/pipermail/ns-developers/2016-April/013524.html and drop tracing at netdevices.
Yes, FlowMonitor needs to be fixed. Can you please post an example showing the issues you found?
Sorry for the late reply. I attach a patch which adds flow monitor to the traffic control example. It can be seen the difference of the throughput shown by the flow monitor and the calculated using GetTotalRx from the Sink application. The difference matches almost exactly with the bytes dropped and requeued.
Created attachment 2430 [details] Add flow monitor to traffic-control example
This bug has been fixed with the revision of the requeue mechanism. Further, I just made the flow monitor able to keep track of the packets dropped by the queue discs. The updated traffic-control example outputs: *** Flow monitor statistics *** Tx Packets: 7682 Tx Bytes: 11509900 Offered Load: 10.003 Mbps Rx Packets: 7641 Rx Bytes: 11448400 Packets Dropped by Queue Disc: 41 Bytes Dropped by Queue Disc: 61500 Packets Dropped by NetDevice: 0 Bytes Dropped by NetDevice: 0 Throughput: 9.94953 Mbps Mean delay: 0.0143646 Mean jitter: 0.00119161 *** Application statistics *** Rx Bytes: 11051136 Average Goodput: 8.84091 Mbit/s *** TC Layer statistics *** Packets dropped by the TC layer: 41 Bytes dropped by the TC layer: 61500 Packets dropped by the netdevice: 0 Packets requeued by the TC layer: 0 The difference in the Rx Bytes is due to the fact that the flow monitor counts the network and transport headers, while the application does not.