Comment 35 for bug 1405549

The problem seems has not been resolved if the controllers have many cpus (as mentioned above).
Right now we have 82 keystone processes running on every controller. If we reboot the controller, the time of getting token is extended up to 4 seconds. We measure this time (using the script above) while the controller is out, and we keep seeing these numbers all the time.
If we set the number of public workers of keystone as 2 on every controller and do the reboot again - we get the time normalized in 30 seconds.
I attach the outputs of the script here.