Comment 3 for bug 1234479

So the 0.2 second difference is another issue.

the _wait_for_power_on call, which is called in a loop. Checks if the power is on each time. And if not re-issues a power on command (https://github.com/openstack/nova/blob/master/nova/virt/baremetal/ipmi.py#L157). So you need to measure the time between power on calls not between power status and power on. It turns out that the difference is a bout 0.5 seconds if not longer.

 I don't know enough about IPMI to say what happens when we keep issuing 'ipmi power on every 0.5 seconds' but that *may* not be good?

Also I don't know what the average time for IPMI to respond is, but I think slowing down the polling interval to 1 second can't hurt. Also we may want to bump up the default value for CONF.baremetal.ipmi_power_retry