Most data centers have limitations on the power and cooling capacity available to support IT equipment. And yet, in many cases capacity is left unused for fear that unexpected peaks in server utilization may trip a circuit breaker. Applying a power cap ensures that your HP ProLiant server or HP BladeSystem enclosure does not use more than the specified number of watts of power and cooling capacity assigned. Because circuits are protected, systems administrators can use power capacity more efficiently and avoid allocating unnecessary power and cooling overhead.

Power is normally allocated by your data center/facilities team based upon the power available within the branch circuit to which the HP ProLiant server or HP BladeSystem enclosure is connected and the cooling capacity of the portion of the data center that will be used. HP recommends setting an initial Dynamic Power Cap by working with facilities to understand available power and cooling capacity and using the HP Power Calculator or HP BladeSystem Sizer to understand how many servers will fit within that capacity. Since these power calculators assume 100% load, sizing numbers will be conservative. Next, use the power management capabilities in Insight Control suite to characterize actual power usage. Based on usage data, work with the facilities organization to re-adjust Dynamic Power Caps as necessary.

Both rack mount servers and server blades within an enclosure normally have a varying power requirement over time. Setting overly aggressive power caps can adversely impact performance. HP recommends capping to the highest observed power usage to avoid impacting server performance.

For HP ProLiant servers, the following components are necessary to make full use of Dynamic Power Capping. These components can be purchased together as part of HP Insight Control suite.

• » iLO Advanced license.
• » Insight Power Manager.

For HP BladeSystems c7000 enclosures, the following components are necessary to make full use of Dynamic Power Capping. These components can be purchased together as part of HP Insight Control suite. These components can be purchased together as part of HP Insight Control suite.

• » iLO Select license.
• » Redundant Onboard Administrator (OA) .
• » Insight Power Manager.
• » N+N Rudundant AC power configuration.

There are numerous characteristics that affect the range of acceptable cap values. These can include hardware configuration, firmware version and servers or other devices that either do not support or have been removed (“opted out”) of the Dynamic Power Cap.

Blades that do not meet the requirements for applying a Dynamic Power Cap (or cannot be communicated with) are treated as unmanageable. Such blades do not prevent enforcement of the cap for the enclosure, but may affect the minimum cap value that can be applied to the enclosure, and the amount of power available to other blades.

In general to improve the range of acceptable cap values you should ensure that all blades are opted-in and that you are running the latest revisions of BIOS, iLO Firmware and OA firmware.

Blade power-on requests are arbitrated by the Dynamic Power Cap software to ensure that circuit protection can be maintained with the addition of a new blade. Although a given blade may be able to fit within the cap during normal operation, Dynamic Power Capping must account for the possible addition of hot-plug of devices, and a potential loss of AC redundancy. If the addition of the new blade will over-subscribe the circuit under these conditions, the Onboard Administrator will not allow the new blade to power on.

Ensuring that all blades within the enclosure can participate in the Dynamic Power Cap will reduce the need to include overhead in the power cap range calculation. It may also be possible to enable more blades to power on by raising the power cap. When raising the power cap, it is important to ensure that you will not over-subscribe the circuit.

The values that affect the cap range are calculated from numerous data items–some of which may vary over time. Configuration changes may also affect the calculation of a minimum cap. Depending upon the configuration changes, a previously acceptable cap value may later become unachievable. Therefore, it is recommended to always apply a cap based upon the constraints of your data center, and not simply pick the minimum possible cap.

Diagnostic information about bays that are unmanageable or opted-out is displayed in the Dynamic Power Cap section of the Power Management configuration page of the Onboard Administrator web interface. Using the command-line interface, the command show enclosure powercap displays this information.

The Peak Power shown in the graphs represents peaks that can be much shorter in duration than would be required to trip a circuit breaker. The Peak Power graphs show the largest 1/2-second average input power during the time interval. Dynamic Power Capping is designed to protect enterprise-class circuit breakers, like those found in HP PDUs, that are designed not to trip unless overloaded by 150% or more for 3 seconds or more. A single 1/2-second peak above the cap in each time interval would appear on the graph as a continuous red area above the cap. However such a series of 1/2-seconds peaks would not threaten a circuit breaker.

In other cases, when a server reboots, the cap is actually disabled while the server re-measures its maximum power. This maximum power could exceed the cap for several seconds. Blade servers under an enclosure-level Dynamic Power Cap request the OA to raise their cap before they reboot. Care must be taken with ML or DL servers with manually set caps not to reboot them all at the same time. For example, "Automatic Power-On Delay" should be set for "Random."

Power caps are enforced by comparing a real-time measurement of the server's power to the cap. In most cases, hot-adding or -removing disk drives or power supplies will have no effect on the server's ability to meet a power cap. However, because adding drives will increase the total power consumption of the server, a server that normally never hits the power cap before the hot-add might hit the power cap (affecting performance) after the hot-add. Also, if a cap is set extremely low and disk drives are hot-added, there is an increased risk that the cap might become unachievable, or that capping to that level would prevent the OS from functioning properly. Hot-swapping power supplies on an ML or DL server could change the combined efficiency of the power supplies, which may shift the server power consumption slightly relative to the cap. This can also have a similar effect on the server as described above for hot-adding disk drives. HP recommends that the server be rebooted soon after a hot-swap event, so the power measurement and capping circuitry can adjust to the new load. Caps that have been set manually should be re-evaluated after a hot-swap event, to verify that they will not adversely affect performance.