Hi, In an OVS 2.0.0 setup, where the data path is loaded with artificially generated traffic by the clients, we observe the effect of different routing algorithms on packet latency. The problem is, despite there exists manually introduced hotspots (that is, switches under excessive traffic load) in the network, change in latency observations due to varying routing algorithms is almost negligible. Considering that the controllers poll the link traffic every 10 seconds and some routing algorithms do not even consider the link traffics, this result really made me thought that maybe OVS is so busy with packet forwarding to the point of becoming the major cause of the latency, regardless of the routing algorithm used. Further, when I check the system resource usage, vswitchd consumes ~550% of the CPU and the controller consumes the rest. (Machine has 6 physical cores.) Which also ensures that the controller is not the bottleneck. What are your thoughts? Am I right in suspecting the OVS as the major cause of the latency? Is there a more accurate way/tool to observe this effect? How should I configure the network to produce more reasonable results for the effect of routing on the latency? Any comments will be really appreciated.
Best. P.S. I repeated the experiment with multiple traffic polling periods (10s, 30s, 60s) and varying routing algorithms (HopCount, Minimax, ShortestPath, etc.), where some routing algorithms (e.g., HopCount) do not even take link traffics into account.
_______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
