switch event generate on ONOS1 by connecting one ovs switch to it by "ovs-vsctl set-controller";
record open flow port status msg by tshark, device and graph event timestamps to calculate differences. In a multi-node test, the node with latest Graph event timestamp is used.
test is run in multiple iterations to obtain statistical results
a number of warm-up runs are used before taking test.
Result:
1node
3node
5node
7node
OFP Port Status -> Device update
2.71
4.14
4.43
4.57
Device update -> Graph update
7.29
15.15
15.14
15.26
1node
3node
5node
7node
OFP Port Status -> Device update
2.43
3.57
3.29
4.14
Device update -> Graph update
1.14
1.29
1.08
1.43
Result Discussions:
"Port up" latency is higher than "Port down". This is largely due to additional ONOS operations in a port-up event. In a port-up event, additional PacketIn/Out messages for link discovery are handled by ONOS, whereas in port-down event, there is no such additional operation.
A typical port-up timing breakdowns for a 3-node cluster test are the following:
OFP of_port_status -> complete Packet In/Out for link discovery: 11ms
OFP of_port_status -> Device Event: 5~11ms
complete Packet In/Out for link discovery -> Link Event: 5~6ms