HP Insight Control Software for Linux - Manual (Page 55)

Various HP ProLiant servers support power capping (which limits the thermal output), and Dynamic Power
Capping (which limits the electrical power consumption). If you are attempting to reclaim thermal capacity,
or limit the total heat output of a group of systems, then either type of cap may be used. If, however, you
are attempting to prevent an electrical overload, be sure that your device supports dynamic power capping.
Similarly, for example, consider a system say ProLiant ML370 server as shown below:
While the faceplate power is 1000 Watts, the Calibrated Max Power as measured by the hardware itself
is only 509 Watts, and based upon historical observation of several months, the highest power consumption
ever observed was 432 Watts. Based on past behavior, there is 502-432 or 70 Watts that can potentially
be reclaimed without noticeably impacting the performance of the system.
When determining an appropriate cap to reclaim capacity, consider the following issues:
1. Do I have sufficient power history collected for this system to make reasonable projections for future
behavior? Several months of power history will likely ensure that you are seeing a typical depiction.
However, if your computing demand has a seasonal component (such as the holidays, tax season, or
end of the fiscal year spikes), you may want to analyze power history covering such time periods, or
allot additional headroom to avoid potential performance impacts of the cap.
2. Applying a Dynamic Power Cap value allows sharing of the cap value dynamically from systems that
are idle to systems that are experiencing demand. But, a cap value applied to a single system does not
adjust automatically, so it is important to ensure that there is sufficient headroom to avoid bumping into
the cap when there is more demand in the future.
3. When considering the requirements for the future, what is the trend of my CPU utilization and therefore
power consumption and thermal output? Are my workloads relatively stabile over time, or is there a
growing need for additional CPU? These factors dictate how much headroom you must leave to avoid
unnecessarily restricting performance.
4. Do I anticipate adding additional hardware (systems in an enclosure or memory, disks, or add-in cards
to a system)? Such additions will increase the upper end of the operating range of the equipment.
5. Consider the business importance of the performance of an individual system before apply any cap
that is significantly below the Calibrated Max Power. If the response time of the system is critically
important, you must apply caps to other systems and ensure that your critical systems always have
access to the absolute highest performance they can deliver.
Task: Reclaiming power/thermal capacity for a new system
As you deploy the new system in your data center, you must select a location that has following sufficient
resources:
•
Physical space in a rack to mount the system
•
Power delivery capacity to plug it into
•
Thermal capacity, or sufficient cooling, to avoid overheating it and surrounding systems
•
Network connectivity of the proper type and bandwidth
•
Location with other related systems and other organizational issues
Task: Reclaiming power/thermal capacity for a new system 55