Our current build system utilization seems to be low in the CPU usage area and our projects builds are typicaly just below 60 minutes.

This is mainly caused by two factors:

1. our maven builds are not parallel in the mvn -T1C sense, and
2. our CSIT test suite has sequential execution

Since we run a ton of different plugins, each executing in a single thread – so we end up with awkward pauses where git-plugin does something, then we check the license, then we run checkstyle...

A quick run of 'mvn clean install -Pq` with or without threading shows some 5x improvement IIRC, which would be on par with this blog. But I got complaints resulting in ODLPARENT-220.

SFT is the next obvious step, as those will cause failures due to concurrent Karaf invocation. Perhaps this could be solved through a launcher plugin, which would allocate Karaf instances using global JVM state.

We obviously need to run some unit tests downstream as well, to see in what shape they are. I suspect controller, netconf and bgpcep will be problematic.