White Paper - Power Management in Intel Architecture Servers

NM works as a simple feedback system, as shown in Figure 6, in which
the user specifies a maximum power limit during a given time period.
The system-level power supply, using the PMBUS standard, measures
and reports the power draw into the system. If it exceeds the specified
limits during a given operation time, a feedback loop that uses ACPI
Source Language code provided by BIOS raises the interrupts to the
OS power management, and an ACPI-enabled OS puts the processor
in lower P-states, and/or t-states, and impact other components such
that the overall system power is reduced.
If system power can’t be brought into the specified limits, a user-
specified action along with the power policy is implemented. This
action may be to raise an alarm (that is, through a SNMP trap to a
management agent) or, in an extreme case, the user can even choose
to shut down the system. The latter may be needed, for example, to
keep a group of systems within the rack-level system limits to avoid
a circuit-breaker trip of the entire rack. Another desired usage is to
prolong the uptime of a server running on a back-up supply if the main
power supply goes down, as often happens in emerging markets.
Node Manager - How it Works
NM works as a Closed Control Loop System
CPU Console Manager
BMCNode Manger
Instrumented
PSUs
*bmc polling
BIOS
(SCI Handler
or ASL code)
Change P/T state from OS
Power Comsumption Set Power Budget
Change
Power
Comsumption
Change number of
P/T-states available
Policy
IPMI or
WSMAN
Figure 6. Node Manager control flow�
8
White Paper: Power Management in Intel® Architecture Servers