HP Integrity NonStop NB54000c Introduction Manual page 18

which is high in the dynamic range of the servers. So the effects of configuration and environmental changes are
ignored. Table 4 shows a summary of the power consumption and capacity savings for the server group.
Table 4. Power consumption for eight DL380 G5 servers when capping to peak power consumption using basic Power Capping
Description
Maximum possible power consumption for 8 DL380 G5 servers
(based on HP Power Calculator)
Maximum power consumption when capping to peak
Savings in power capacity
Additional servers that can be provisioned within the same cooling infrastructure
If all eight servers supported Dynamic Power Capping, administrators could use the 3116-watt power cap to
manage electrical provisioning rather than provisioning to the faceplate value of the power supply. Table 5
illustrates how many more servers the same circuit could power under the 3116-watt power cap. Derating the
faceplate value by 20 percent requires 7760 watts. This example reserves a 10 percent guardband for other factors
(see "Providing a guardband for a power capping group"). With Dynamic Power Capping applied, the same
circuit can support over 10 more servers, if all 10 have the same power cap as the first eight servers.
Table 5. Power provisioning for eight DL380 G5 servers when capping to peak power consumption using Dynamic Power
Capping
Description
Traditional input power provisioning for 8 DL380 G5 servers
(1217 input VA at 208 volts de-rated by 20 percent)
Maximum power consumption when capping to peak
Guardband
Savings in power capacity
Additional servers that can be provisioned within same 7760 watts using Dynamic Power
Capping when capping to peak
It is important to understand this: Capping to peak based on the power consumption graph for a server group is
actually capping to the peak of the total power consumptions of servers in the group. It assumes that all the servers
in the group consume the same amount of power. This would not be true under these conditions:
If the servers were configured differently
•
• If workloads on individual servers peaked at different times
If workloads varied significantly from server to server
•
If all servers in the group did not consume the same amount of power, the cap in this example would have an
adverse affect on server performance, especially if they support Dynamic Power Capping. If the servers in a group
Power
3384 watts
3116 watts
268 watts
0.7
Power
7760 watts
(970 x 8)
3116 watts
312 watts
4332 watts
10.1
18