HP ProLiant DL280 HP Power Capping and HP Dynamic Power Capping for ProLiant s - Page 23

Peak power reporting and Dynamic Power Capping

Page 23 highlights

Peak power reporting and Dynamic Power Capping Both iLO and Insight Control report the power metrics for ProLiant servers, including peak power consumption. The power monitoring system for servers records peak power consumption as the largest half-second power measurement in a 5-minute period. It can take up to one-half second to bring a server's power consumption back below the cap. As a result, it is possible to see reported peak power numbers that are several watts above a server's power cap value. Dynamic Power Capping still provides circuit protection, since it will always bring server power consumption below the cap long before a circuit breaker can trip. Using HP Power Regulator in conjunction with power capping Power capping and HP Power Regulator use the same underlying power management system to manage and control server power consumption. However, HP designed these two power management features to accomplish different goals. Power capping provides strict control of a server's maximum power consumption. It can limit a server's maximum power consumption over a wider range of power than Power Regulator can, without regard for the affect on system performance. Power capping can use both processor P-states and clock throttling to limit system power consumption. Power Regulator focuses on optimizing power use in a server (or group of servers) without affecting server performance. In its recommended configuration, Power Regulator dynamically selects the most power-efficient processor P-state that matches the present workload of the server. By doing this, Power Regulator lowers power consumption without affecting system performance or throughput. In this sense, Power Regulator is primarily an efficiency tool. Administrators can use power capping and HP Power Regulator on an individual server or server group at the same time without creating a conflict. Power Regulator manages server power use as long as a server's power consumption remains below the power cap setting. Power capping will override Power Regulator only if power consumption exceeds the established power cap. Power capping and CPU utilization Both power capping and Power Regulator involve manipulating the system processor at a very basic level. As a result, power capping can affect the processor utilization that monitoring tools report at the operating system level. If a server with a relatively constant workload has a power cap low enough to engage the power control mechanisms, OS monitoring utilities will generally report a greater CPU utilization number than they would if the cap were not in place. Power capping and option cards The server's capping circuit indirectly controls option cards installed in PCI Express slots. That is, as the capping circuit decreases to keep the server power below the cap, the processor is less able to issue commands to the card. Option cards that consume much more power during normal operation than during the maximum-power test at POST complicate identifying the server's maximum and minimum power consumption and the minimum capping threshold. As best practice, allow an application that includes such option cards to run several minutes before choosing a cap value so iLO can measure the true application power. Providing a guardband for a power capping group The sophisticated algorithms used by Enclosure Dynamic Power Capping automatically manage all the conditions that can affect the total enclosure power consumption. However, they do not automatically manage individual servers singly through iLO or as groups through Insight Control. You must manually manage other factors that are beyond the control of the individual server power capping subsystem. After considering all of these factors, you should set aside some percentage of the available power as a reserve or guardband. The total amount of guardband that is appropriate for any installation will depend on the servers and the size and behavior of the environment of that installation. 23

  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25

23
Peak power reporting and Dynamic Power Capping
Both iLO and Insight Control report the power metrics for ProLiant servers, including peak power consumption. The
power monitoring system for servers records peak power consumption as the largest half-second power
measurement in a 5-minute period. It can take up to one-half second to bring a server’s power consumption back
below the cap. As a result, it is possible to see reported peak power numbers that are several watts above a
server’s power cap value. Dynamic Power Capping still provides circuit protection, since it will always bring server
power consumption below the cap long before a circuit breaker can trip.
Using HP Power Regulator in conjunction with power capping
Power capping and HP Power Regulator use the same underlying power management system to manage and
control server power consumption. However, HP designed these two power management features to accomplish
different goals.
Power capping provides strict control of a server’s maximum power consumption. It can limit a server’s maximum
power consumption over a wider range of power than Power Regulator can, without regard for the affect on system
performance. Power capping can use both processor P-states and clock throttling to limit system power consumption.
Power Regulator focuses on optimizing power use in a server (or group of servers) without affecting server
performance. In its recommended configuration, Power Regulator dynamically selects the most power-efficient
processor P-state that matches the present workload of the server. By doing this, Power Regulator lowers power
consumption without affecting system performance or throughput. In this sense, Power Regulator is primarily an
efficiency tool.
Administrators can use power capping and HP Power Regulator on an individual server or server group at the same
time without creating a conflict. Power Regulator manages server power use as long as a server’s power
consumption remains below the power cap setting. Power capping will override Power Regulator only if power
consumption exceeds the established power cap.
Power capping and CPU utilization
Both power capping and Power Regulator involve manipulating the system processor at a very basic level. As a
result, power capping can affect the processor utilization that monitoring tools report at the operating system level. If
a server with a relatively constant workload has a power cap low enough to engage the power control mechanisms,
OS monitoring utilities will generally report a greater CPU utilization number than they would if the cap were not in
place.
Power capping and option cards
The server’s capping circuit indirectly controls option cards installed in PCI Express slots. That is, as the capping
circuit decreases to keep the server power below the cap, the processor is less able to issue commands to the card.
Option cards that consume much more power during normal operation than during the maximum-power test at
POST complicate identifying the server’s maximum and minimum power consumption and the minimum capping
threshold. As best practice, allow an application that includes such option cards to run several minutes before
choosing a cap value so iLO can measure the true application power.
Providing a guardband for a power capping group
The sophisticated algorithms used by Enclosure Dynamic Power Capping automatically manage all the conditions
that can affect the total enclosure power consumption. However, they do not automatically manage individual
servers singly through iLO or as groups through Insight Control. You must manually manage other factors that are
beyond the control of the individual server power capping subsystem. After considering all of these factors, you
should set aside some percentage of the available power as a reserve or guardband. The total amount of
guardband that is appropriate for any installation will depend on the servers and the size and behavior of the
environment of that installation.