Blackbird - SB:Port Event Latency

Reference:

• for Test Plan see: Test Plan - Perf & Scale-out; 
• ONLabTest script - https://github.com/opennetworkinglab/ONLabTest: TopoPerfNextBM

Test Setup:

• RC2 - commit 4964d3d768deb0a8b5773c84f47787db455ceaba
• Using OVS version 2.3.1/OF 1.3
• Bare-metal Servers: dual-Xeon E5-2670, 32GB DDR3 RAM, SSD, Cluster Network is 1Gbps
• JAVA_OPTS="${JAVA_OPTS:--Xms8G -Xmx8G}"
• "org.onosproject.net.topology.impl.DefaultTopologyProvider.cfg": maxEvents = 1; maxIdleMs = 0; maxBatchMs = 0

Test Procedure:

• 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:

Result Discussions:

1. "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.
2.  A typical port-up timing breakdowns for a 3-node cluster test are the following:
   1. OFP of_port_status -> complete Packet In/Out for link discovery: 11ms
   2. OFP of_port_status  -> Device Event: 5~11ms
   3. complete Packet In/Out for link discovery -> Link Event: 5~6ms
   4. Link Event -> Graph Event: 1~2ms