Dynamic Power Capping - HP DL370 - ProLiant - G6 Performance Introduction Manual

The previous generation of ProLiant servers marked a shift away from processors as the primary
producers of heat in the server. As memory modules become denser, they generate more heat. To
combat this, DDR-3 DIMMs, as used in the ProLiant G6 servers, incorporate the first reliable on-DIMM
thermal sensors
Because hard drive thermal sensors were not directly associated with fans, the fans would often
operate at high speeds to prevent hard drives from overheating. ProLiant 300-series G6 servers
incorporate hard drive temperature sensors into the body of data used to determine fan speed. This
requires collaboration among various pieces of firmware, including the iLO 2 firmware, system
firmware, and RAID storage controller firmware. The 300-series G6 servers have "zoned" fans that
increase cooling and energy efficiencies in the server by adjusting cooling to those zones when called
for by the sensors. This provides improved efficiency and better acoustics for the platform. The iLO 2
management processor in the G6 300 series uses a sophisticated control algorithm to set the speed
for each fan zone in the system based on feedback from the appropriate temperature sensors. This
allows fans to consume the minimum amount of required power.
The fan control algorithm lets ProLiant 300-series G6 servers change fan speed as the situation
dictates. In ProLiant servers prior to G6, if one fan failed, all the other fans were set to high speed to
assure the server remained within thermal specifications. ProLiant 300-series G6 servers now include
enough sensors to construct an accurate view of the thermal landscape within the server, allowing the
sensors and the fan control algorithm to determine if fan speeds need to be increased.
An RBSU option called "increased cooling" is included in the ProLiant 300-series G6 servers. The user
can select this option and manually drive fans to higher speeds for use during situations requiring
additional cooling.

Dynamic Power Capping

All ProLiant 300-series G6 servers incorporate Dynamic Power Capping, which allows administrators
to cap power use and more accurately allocate the amount of power a server will use. Dynamic
Power Capping can bring a server experiencing a sudden increase in workload back under its power
cap in less than one-half second, preventing any surge in power demand that could cause a typical
data center circuit breaker to trip. Dynamic Power Capping has been designed and tested (at 50
degrees C and 150 percent overload) to ensure that it can prevent tripping circuit breakers that have
a specified trip time of three seconds or longer.
This ability to keep server power consumption below the power cap in real time means that Dynamic
Power Capping can be used as an effective tool in planning and managing both electrical
provisioning and cooling requirements in the data center. An administrator can electrically provision a
PDU or a rack to less than the full faceplate power rating of all the servers loaded.
To implement Dynamic Power Capping, the iLO 2 management processor works in conjunction with a
power microcontroller both to measure and to control power use. When enforcing the Dynamic Power
Cap, the power microcontroller keeps the processor's performance and power use under the set cap.
This process is illustrated in Figure 6. Dynamic Power Caps for individual servers can be set from the
iLO 2 Advanced user interface. Dynamic Power Caps for multiple rack-mount servers may be set from
the power management module within HP Insight Control Environment.
HP Dynamic Power Capping is operating system independent and functions with all operating systems
and software applications. HP Dynamic Power Capping will continue to function even if the software
fails because it utilizes operating system independent hardware.
Since Dynamic Power Capping can impact server performance if set too aggressively, HP
recommends that Dynamic Power Caps be set at values that match or exceed the highest observed
power consumption over a representative server workload sample.
15