Chapter 2. Architectural and technical overview
67
something that should never be reached but frees up margined
power in the data center. The
margined power
is the amount of extra
power that is allocated to a server during its installation in a data
center. It is based on the server environmental specifications that
usually are never reached. Server specifications are always based on
maximum configurations and worst case scenarios.
Processor core nap
The IBM POWER6 processor uses a low-power mode called
nap
that
stops processor execution when there is no work to do on that
processor core
(both threads are idle). Nap mode allows the
hardware to clock-off most of the circuits inside the processor core.
Reducing active power consumption by turning off the clocks allows
the temperature to fall, which further reduces leakage (static) power
of the circuits causing a cumulative effect. Unlicensed cores are kept
in core nap until they are licensed and return to core nap whenever
they are unlicensed again.
EnergyScale for I/O
IBM POWER6 processor-based systems automatically power off
pluggable, PCI adapter slots that are empty or not being used, saving
approximately 14 watts per slot. System firmware automatically
scans all pluggable PCI slots at regular intervals looking for slots that
meet the criteria of not being in use, and then powers them off. This
support is available for all POWER6 processor-based servers, and
the expansion units that they support. Note that it applies to hot
pluggable PCI slots only.
Oversubscription protection
In systems with dual or redundant power supplies, additional
performance can be obtained by using the combined supply
capabilities of all supplies. However, if one of the supplies fails, the
power management immediately switches to normal or reduced
levels of operation to avoid oversubscribing the functioning power
subsystem. This can also allow less-expensive servers to be built for
a higher (common-case) performance requirement while maintaining
the reliability, availability, and serviceability (RAS) redundancy feature
expected of IBM servers.
System implementations
Although the basic design of the EnergyScale architecture is similar for all of the POWER6
processor-based systems, some differences on system implementations exist.
The Power 595 is the largest POWER6 processor-based server. These servers contain
multiple boards, and the designs of their predecessor in the POWER5 product line already
contain power measurement features. Such machines pose a significant challenge because
of their scale and the additional complexity imposed by their hardware designs. The design
approach for extending the EnergyScale architecture to them involves three changes:
The EnergyScale architecture uses the existing power measurement function provided by
the BPCs used in the power supplies.
Rather than adding a TPMD card to each board, the design uses existing microcontrollers
that are already embedded in the power distribution subsystem (inside DCA assembly).
This allows real-time control on each board.
System-wide changes, such as to the frequency and the reporting of system-wide
measurements, use non-real-time implementations running on a service processor.
Although this limits the responsiveness of the power management system, this allows it to
scale to the scope needed to control a very large machine.
Summary of Contents for Power 595
Page 2: ......
Page 120: ...108 IBM Power 595 Technical Overview and Introduction...
Page 182: ...170 IBM Power 595 Technical Overview and Introduction...
Page 186: ...174 IBM Power 595 Technical Overview and Introduction...
Page 187: ......
