If the BMC had a way to use a private connection to raise a signal (for example, powerfail) on each node, yes. bmcd would need to be modified to send the signal. This is what a smart UPS does (example: APC PowerChute). The challenge is an agent would be needed on each node to trap the signal and execute an immediate shutdown with power off. Nagios and Zabbix have these capabilities, but would need to be substantially stripped of features to make them lightweight enough to be practical on the BMC. At one time, Redfish was going to be implemented as a REST interface on the BMC. That would probably be lighter weight and not lock into a specific user interface. Redfish is also supposed to be scalable to a manager of managers. The challenge is a management REST agent on each node that the BMC could talk to. Clearly a good use of a private VLAN between the BMC and compute nodes.