50
7X24 MAGAZINE FALL 2014
How is Power Consumed?
Power is consumed in a system in
primarily three ways:
1. Active power known as AC power
from wire and transistor switching
2. Static power known as DC power or
leakage power which has no useful
purpose
3. Power of electronic devices
represented by capacitance x
voltage squared x frequency x active
switching factor.
Methods of Server Power
Management
There has been an overall trend to
increase the ability for advanced
configurations of CPUs and other
devices by the operating system. This
increased ability to control performance
and energy is enabling users to have
more ability to select tradeoffs that best
meet their business needs. More
energy-efficient server technologies
include:
n
More efficient power supplies,
better DC voltage regulators,
processors that consume less power,
and cooling fans that are variable
speed and more energy-efficient.
n
CPU speed throttling and power
saving technology that reduces the
processor's clock rate and voltage
when the processor is idle. The aim is
to minimize overall power
consumption and lower heat
generation, allowing for slower (and
quieter) cooling fan operation.
n
The latest operating systems, which
take advantage of hardware features
to deliver optimal power efficiency
for any given workload running on
the
server.
For
example,
continuously alter the power states
of the processors in the system in
response to the utilization level of
the workload. This ensures that
processor power usage maps to the
needs of the workload, with minimal
impact on performance.
Examples of Server Power
Management Features
The overreaching goal is when
something is not being used to turn it
off. If something (e.g. system, chip, core,
memory, etc.) is heavily utilized then
speed it up to maximize performance,
and if it not fully utilized then slow it
down to save power. Some of the
management features are imbedded
within the system firmware and others
allow the user to configure the system
to optimize the business requirements
(higher performance to meet customer
demands) or the operation of the data
center infrastructure (saving energy,
lower noise, etc.). Here are few
examples (some of the definitions may
change slightly depending on the
manufacturer) in each category focusing
on just the processor.
Server Controlled:
n
Performance States
– A power
regulator is implemented in the
system firmware and directly
monitors the instruction load of the
server processor(s) to determine the
level of system activity. The power
regulator uses this information to
continuously adjust the performance
states of the processor(s) to match
processor power consumption to the
current workload without noticeably
impacting
overall
system
performance.
n
CPU States
– The basic idea of
these modes is to cut the clock
signal and power from idle units
inside the CPU. The more units you
stop (by cutting the clock), reduce
the voltage or even completely shut
down, more energy you save, but
more time is required for the CPU to
“wake up” and be again 100%
operational.
n
Core Parking
– A feature that
consolidates work onto a smaller
number of processor cores when
workload utilization is low. Processor
cores that aren’t doing any work are
placed into a deep sleep state. This
feature effectively scales the number
of processor cores in active use.
n
Dynamic Voltage Frequency
Scaling
– Dynamic voltage and
frequency scaling slows the clock
speed saving active power (linear
reduction of frequency) but more
importantly it allows for voltage
reduction (V2 in the equation) while
still
delivering
predictable
performance.
User Controlled:
n
Frequency Control
– Fixed
frequency control, in some cases,
allows the user to specify a fixed
frequency for all processors in the
system to run. The frequency must
be within the supported frequency
range.
n
Power Capping
– Power capping
enforces a user-specified limit on
power consumption. The main
purpose of the power cap is not to
save power but rather to give the
data center operator the capability
to reallocate power from existing
systems to new systems by reducing
the margin assigned to the existing
servers. It gives the operator the
capability to add extra servers to a
data center which previously had all
available power allocated to its
existing systems.
Although the examples above are for
processors,
power
management
features can be deployed for other
components. For most systems memory
can also use a large amount of energy
and is a good candidate for power
management. Since frequency and
voltage cannot be dynamically adjusted
for memory only, data throughput is
gated. Other components within the
system can be adjusted depending on
their features that allow adjustments
within their range of operation and in
support of the memory and processor
workloads.
Is Server Reliability
Affected by Energy
Management Features?
Obviously, server reliability is another
knob that the server manufacturer can
use when affecting various energy
