Comment 17 for bug 1250168

For #1 I can pass the test with 150 vms locally (a 3GB ram VM with 2 vcpu running on macos) in 250 seconds.

For #2
Dims has a patch which albeit not resolutive improves nova/neutron communication by cutting off keystone chatter.
We can do a blueprint and call it make-neutron-not-slow-as-molasses, but it will target nova :) Also, molasses don't move anyway... so perhaps you might want to use a snail!

Jokes apart, there are three aspects to this task:
1 - reduction of chatter between nova and neutron (either critical bug or blueprint, but I'd prefer the first as it gives sense of urgency and does not really add a new feature/capability or performs extensive code changes)

This one is possibly the most important from the perspective of large-ops (and possibly parallel) tests.

2 - move out network allocation from the compute node to the API node (There has already been a lot of discussion on the mailing list); this is a blueprint, which I think arosen already registered some time ago - it needs discussion and smoothing out of details especially wrt baremetal, but it would be good to land it in icehouse.
3 - reduce contention in the neutron server, especially around IPAM. This is another delicate task as it might involve major changes in Neutron IPAM logic.