Power & Cooling Policies

Beyond basic power on/off controls, the BMC provides sophisticated tools for managing the server's power consumption and thermal performance. These features are designed for data center environments where balancing performance, power usage, and cooling is critical. By setting power policies, you can ensure your servers operate within specific energy budgets without sacrificing performance where it's needed most.

This chapter introduces the Node Manager and CUPS features for advanced power and performance tuning.

Using the Node Manager to set power policies

Node Manager is an Intel technology that provides platform-level power monitoring and policy enforcement. It allows you to view detailed power usage statistics and create rules to limit power consumption for specific domains like the CPU or the entire system.

To view power statistics:

1. In the sidebar menu, navigate to Settings > Node Manager.

2. The Power Statistics tab shows live and historical power usage (in watts) and throttling information for various system components.

[Image, EXISTING, Source: 7.13: Screenshot of the Node Manager power statistics page, showing a table of domains, current, average, and maximum power values.]

To create a power policy:

1. Select the NM Configuration tab.

2. Click Create New Policy.

   
3. In the popup window, define the policy parameters:

   • Domain: The component you want to control (e.g., CPU Subsystem, Memory Subsystem).

   • Policy Limit: The maximum power (in watts) you want to allow for this domain.

   • Policy Storage: Set to persistent to ensure the policy remains active after a reboot.

4. Click Create New Policy to save and apply the rule.

[Image, EXISTING, Source: 7.13: Screenshot of the "Create New Policy" dialog, showing fields for domain, policy limit, trigger type, etc.]

Using CUPS to configure workload-based performance

CUPS (Configurable Usage Policy Service) is a feature that tunes the server's power and thermal management based on its expected workload. It helps you optimize the system for specific roles, such as a CPU-intensive compute node or a memory-intensive database server.

1. Navigate to Settings > CUPS.

2. Choose a configuration mode:

   • Static: Manually assign a load factor percentage to the Core Load Factor, Memory Load Factor, and I/O Load Factor. The sum of the three factors must equal 100. This is best for predictable, unchanging workloads.

   • Dynamic: Allows the system to automatically adjust the load factors based on real-time usage. This is better for variable workloads.

3. Click Save settings.

[Image, EXISTING, Source: 7.12: Screenshot of the CUPS configuration page, with fields for static and dynamic load factors.]

Best Practice: Start with monitoring, then tune

Before creating aggressive power policies, it's a good practice to first use the Node Manager to monitor your server's power consumption under a typical workload. This gives you a baseline to understand its actual power needs. Only after you have this data should you start applying power limits to avoid unnecessarily throttling performance. For most general-purpose servers, leaving CUPS in Dynamic mode is the recommended approach.

Field Reference (Node Manager Policy)

Field	Description
Domain	The hardware area the policy will control (e.g., CPU, DRAM, Platform).
Policy ID	A custom identifier for the policy (numeric or string).
Component id	A unique integer ID to bind the policy to a specific hardware component.
Policy Limit	The power limit in watts. This must be within the minimum and maximum capability values of the selected domain.
Limit Exception	Action to take if the policy limit cannot be enforced.
State	The current state of the policy (Enabled or Disabled).
policy Correction Time (ms)	The duration (1000-60000 ms) over which the power correction is applied.
Statistics Reporting Period	How frequently the policy's metrics are reported (in seconds or minutes).
Trigger Type	The event that activates the policy (e.g., AlwaysOn, OnTimer).
Trigger Limit	A threshold value that must be met for the trigger to activate.
Power Correction Type	The method used to enforce the power limit.
Policy Storage	Defines whether the policy is temporary or persistent across reboots.