Dynamic Power Allocation for Information Handling Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional power capping in data centers leads to trapped power, where some servers are power-constrained while others are idle, limiting the number of servers that can operate at their rated power, resulting in wasted potential work due to inefficient power allocation.
Innovation Solution
Implementing a dynamic power allocation system where information handling system nodes can share power based on real-time usage, allowing nodes to borrow or lend power credits to optimize power usage within a group, ensuring that the total power limit is not exceeded and minimizing trapped power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If conventional power capping is implemented to limit total power consumption, then power consumption is controlled within limits, but trapped power occurs where some servers are power-constrained while others are idle
Solution Approach 1:
The patent implements dynamic power allocation where power caps are adjusted in real-time based on actual power consumption and workload demands. Instead of static power caps, the system continuously monitors power usage and dynamically reallocates power credits between servers, allowing power-constrained servers to borrow power when others are idle, thereby eliminating trapped power while maintaining total power limits
Solution Approach 2:
The system employs feedback mechanisms where power consumption data from each server is continuously monitored and fed back to the power management controller. This feedback enables the system to identify idle servers with unused power capacity and redirect their excess power to servers that are power-constrained, optimizing overall power utilization and eliminating trapped power
2Loss of energy
If static power caps are assigned to each server, then total power consumption is controlled, but power allocation efficiency decreases due to idle power capacity
Solution Approach 1:
The patent introduces a power management controller as an intermediary between servers and power sources. This controller acts as a mediator that manages power credits and facilitates real-time power reallocation between servers. The intermediary handles the complexity of dynamic power management, coordinating power transfers and ensuring total power limits are maintained while optimizing power distribution efficiency
3Quantity of substance
If power is capped at individual server level, then power limits are enforced, but the number of servers that can operate at rated power is limited
Solution Approach 1:
The patent merges power capacity across multiple servers into a shared power pool. Instead of treating each server's power capacity independently, the system combines available power capacity from all servers and dynamically distributes it based on real-time needs. This merging allows more servers to operate at their rated power by sharing the total power capacity, effectively increasing the number of operational servers without exceeding total power limits
Data Source
AI summary
Systems and methods are disclosed that may be implemented to dynamically allocate relative power consumption between a group of multiple information handling system nodes that share a common (e.g., capacity-limited) power supply or source of power. The relative power consumption of the multiple information handling system nodes may be adjusted based on real time power consumption of each of the individual information handling system nodes, as well as the need for additional power by one or more of the individual information handling system nodes. A group of multiple information handling system nodes may dynamically communicate power usage characteristics in a distributed manner between themselves to implement a peer-to-peer acknowledgement architecture, or alternatively may communicate power usage characteristics to a centralized power manager.