Server Power Allocation Renewal for Data Center Energy Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Managing power efficiently across a large number of servers in a datacenter is challenging, particularly for external agents trying to dynamically adjust power consumption to avoid exceeding power caps, as frequent polling is required to ensure effective power allocation, and servers may not receive timely renewals or communications due to network issues.
Innovation Solution
Implementing a group power management entity that allocates and renews power to servers within a defined budget, allowing servers to autonomously reduce power consumption if allocations are not renewed, using local controllers to manage power states and negotiate power allocations with the entity, ensuring power is allocated based on workload and available budget.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If frequent polling is used to dynamically adjust power consumption, then power allocation effectiveness is improved, but network communication overhead and system complexity increase
Solution Approach 1:
The patent applies preliminary action by having servers proactively request power allocations before they are needed, rather than waiting for external polling. The server power management entity grants power allocations with specified durations in advance, allowing servers to plan their power consumption without requiring frequent polling checks. This shifts the initiative from reactive polling to proactive power allocation requests.
Solution Approach 2:
The patent implements self-service by enabling servers to autonomously manage their own power allocations by requesting and renewing power from the server power management entity without external intervention. Each server independently monitors its own power allocation status and initiates renewal requests, eliminating the need for external agents to poll each server individually and reducing network communication overhead.
2Device complexity
If power allocations are granted for extended durations, then network communication frequency is reduced, but responsiveness to power cap changes decreases
Solution Approach 1:
The patent applies dynamics by making power allocation durations variable rather than fixed. The server power management entity can adjust the duration of power allocations dynamically based on current power cap conditions, workload requirements, and server performance. This allows the system to grant longer durations when power availability is stable and shorter durations when rapid adjustments are needed, balancing communication reduction with responsiveness.
Solution Approach 2:
The patent implements feedback through continuous monitoring of power consumption, allocation status, and power cap conditions. The server power management entity receives feedback from servers about their power usage and allocation status, and adjusts future power allocation decisions accordingly. This feedback loop ensures that even with extended allocation durations, the system can respond to changing power cap conditions by adjusting the timing and duration of subsequent allocations.
3Loss of energy
If servers autonomously reduce power consumption, then energy efficiency is improved, but risk of premature power reduction increases
Solution Approach 1:
The patent applies beforehand cushioning by implementing a buffer mechanism where servers can request power allocation renewals before their current allocations expire. The server power management entity can grant overlapping or extended allocations to prevent premature expiration, cushioning against network delays or communication failures that might otherwise cause unintended power reductions. This ensures continuous power supply even during transient communication issues.
Data Source
AI summary
One embodiment provides a power management method for servers in a data center. A group of servers is selected, and the total power allocated to a group of servers is limited to within a group power budget. A separate server power allocation is individually requested for each of a plurality of the servers. Within the constraints of the group power budget, the requested server power allocations are selectively granted for a specified magnitude and duration. The granted server power allocations are also selectively renewed, either automatically or upon request of the servers. Each server that has not received a renewed server power allocation from a group power management entity upon the expiration of the specified duration automatically reduces its own power consumption, such as by the server powering itself off.