I ran a few tests yesterday to see how much the latest flattening [1] efforts effect performance on the TripleO Heat templates. I only ran a few of each type but the results looked towards a minor but positive improvement in the speed with which the flattened templates get processed via Heat.
Using an all-in-one standalone TripleO deployment I timed how long it took for 'openstack tripleo deploy --standalone --output-only' to run using the latest Heat binaries. I then compared the results switching between t-h-t from master and rocky. On my laptop the results pointed to about a 30-45 second improvement in wall time of 5 minutes or so. A reasonably minor improvement but it is definately in the right direction. The more complex number of services and roles you have the greater this number is likely to be.
I don't think performance was the primary reason we undertook this restructuring but I had several people ask about potential performance improvements so I wanted to provide some feedback here.
Disclosure: In order to test rocky I had to make a few minor changes to t-h-t in order to get the latest all-in-one to run that similarly. Basically just backporting tools/process_templates.py and removing a few invalid network variable references. Here is the branch [2] if you are interested.
Dan
[1] https://blueprints.launchpad.net/tripleo/+spec/services-yaml-flattening [2] https://github.com/dprince/tripleo-heat-templates/tree/rocky_test