24
Administrators should consider the following factors for determining the amount of power to reserve as a
guardband:
•
The guardband should allow any individual server in the group to power up at full power while constraining
other servers on the same circuit to their cap value. The preceding “Powering up groups of servers when using
Dynamic Power Capping” section illustrates this.
•
If only a few servers share a circuit, administrators should apply additional guardband equal to the worst-case
tolerance of the server power meters. However, meter errors can be both positive and negative, so guardband
for meter tolerance, as a percentage of the cap, can approach zero as the number of servers increases.
•
Hardware and environmental changes can increase server power consumption and cause a power cap to affect
performance. In the extreme, the power cap can become unachievable, as described in “Setting low or
unattainable power caps on servers.” Administrators expecting to make significant hardware or environmental
changes for one or more servers should increase guardband to the group total to allow for the changes.
However, best practice is to re-evaluate the server caps after adding hardware and to select caps that are at
least halfway between the minimum and maximum power consumption for a server or server group. When you
follow best practices, additional guardband at the group level is generally not necessary.
Tables 5 and 6 show examples of applying a guardband to a group of servers.
Summary
HP Power Capping and HP Dynamic Power Capping are important power management features of HP ProLiant
servers. System administrators can use power capping to set limits on power consumption for a single server or
groups of servers that do not affect server performance. Administrators can provision data center power and cooling
resources to appropriate levels rather than to unrealistic maximum requirements. Enclosure Dynamic Power Capping
is the newest and most sophisticated implementation of power capping. It can actively reallocate an enclosure’s
power cap across individual server blades based on their workload requirements.
