Rack-Level Power Scheduling for Data Center Power Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face inefficiencies due to underutilization of computing resources and excessive power usage, leading to increased carbon footprint and reduced efficiency, as existing scheduling methods focus on individual server power rather than rack-level capacity.
Innovation Solution
Implementing a power-aware scheduler that allocates processes based on expected power consumption values and available rack power capacity, optimizing server rack utilization by scheduling on racks with the largest available power capacity to avoid exceeding peak power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If scheduling is based on individual server power rather than rack-level capacity, then individual server power management is simplified, but overall data center power efficiency deteriorates and rack power capacity is underutilized
Solution Approach 1:
The patent transitions from one-dimensional individual server power management to two-dimensional rack-level power capacity management. The scheduler considers both individual server power consumption and the aggregate rack power capacity, allocating workloads across multiple dimensions to optimize overall power efficiency while maintaining individual server manageability.
Solution Approach 2:
The patent merges individual server power management with rack-level power capacity management into a unified scheduling approach. By combining these two levels of power management, the system achieves both simplified individual server operation and optimized overall data center power efficiency, resolving the contradiction between ease of operation and energy efficiency.
2Reliability
If fixed power overhead is maintained for safety margins, then mission-critical reliability is ensured, but computing device utilization deteriorates and efficiency is reduced
Solution Approach 1:
The patent implements dynamic power overhead adjustment based on real-time rack power capacity and workload characteristics. Instead of maintaining static fixed power overhead, the system dynamically adapts the safety margin according to current conditions, allowing computing devices to operate at higher utilization while maintaining reliability through adaptive rather than rigid safety mechanisms.
Solution Approach 2:
The patent changes the power overhead parameter from a fixed value to a variable that adjusts based on rack power capacity, workload type, and system conditions. This parameter change enables the system to maintain mission-critical reliability when needed while maximizing computing device utilization during periods when safety margins can be reduced, resolving the contradiction between reliability and productivity.
3Device complexity
If rack power capacity is not monitored and utilized, then system complexity is reduced, but power usage efficiency deteriorates and carbon footprint increases
Solution Approach 1:
The patent implements feedback mechanisms that monitor rack power capacity and use this information to optimize workload scheduling. The scheduler receives feedback about available rack power capacity and adjusts workload allocation accordingly, creating a closed-loop system that improves power usage efficiency without requiring overly complex manual monitoring, as the feedback is automatically processed by the scheduling algorithm.
Solution Approach 2:
The patent enables the scheduling system to automatically manage rack power capacity utilization without requiring external intervention or complex manual oversight. The scheduler self-adjusts workload allocation based on monitored power capacity, allowing the system to serve itself in optimizing power efficiency while maintaining manageable complexity through automated decision-making algorithms.
Data Source
AI summary
A computing device determines an expected power consumption value for a set of processes. The computing device further determines an available rack power capacity for one or more server racks amongst multiple server racks, and selects a first server rack having a highest available rack power capacity. The first server rack includes a first set of servers, and the computing device further determines whether a first server of the first set of servers is available. Responsive to determining the first server is available, the computing devices assigns the set of processes to the first server.