Predictive Power Management for Multi-Phase Datacenter UPS
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current power management systems in datacenters, such as VMware DPM and Eaton's Intelligent Power Manager, fail to effectively predict and optimize power consumption during power events or grid instability, leading to potential degradation of phase balance and increased energy costs, especially during business continuity plans and energy demand response mechanisms.
Innovation Solution
A method and system utilizing a predictive model and machine learning algorithms to process action inputs impacting power consumption, optimizing server utilization based on battery autonomy and load balancing of multiple phase power supplies, which determines sequences of shutdown or shifting actions for virtual machines and physical servers to manage power effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If VMs are migrated and consolidated on fewer physical servers to optimize power consumption, then energy efficiency is improved, but the system's ability to respond to power events and maintain phase balance deteriorates
Solution Approach 1:
The system pre-calculates and stores migration plans that consider multiple scenarios including power events and phase balance requirements. When a power event occurs, the pre-computed plans enable rapid response without compromising phase balance, resolving the contradiction between energy optimization and adaptability to power events.
Solution Approach 2:
The power management system dynamically adjusts VM migration decisions based on real-time power events and grid conditions. Instead of static consolidation, the system continuously adapts server utilization to maintain phase balance while optimizing power consumption, allowing the system to respond flexibly to changing power conditions.
2Duration of action of moving object
If shutdown actions are taken during power events to extend battery autonomy, then UPS autonomy is improved, but service continuity deteriorates
Solution Approach 1:
The system pre-computes optimized shutdown sequences that prioritize non-critical VMs while preserving critical services. When power events occur, these pre-planned sequences enable the system to extend battery autonomy through intelligent shutdown decisions rather than arbitrary or reactive shutdowns, maintaining service continuity for critical functions.
Solution Approach 2:
The system applies different shutdown strategies to different VMs based on their criticality and power consumption characteristics. Critical VMs are protected and kept running longer, while non-critical VMs are shut down to extend battery autonomy, resolving the contradiction between extending battery life and maintaining service continuity through differentiated local decisions.
3Loss of energy
If aggressive load shedding is implemented to reduce power consumption during grid instability, then energy costs are reduced, but phase balance deteriorates
Solution Approach 1:
The system continuously monitors phase balance and power consumption, using this feedback to adjust load shedding decisions in real-time. When phase imbalance is detected, the system modifies migration and shutdown actions to restore balance, enabling the system to reduce energy costs while maintaining stable phase composition through continuous adaptive control.
Solution Approach 2:
The load shedding strategy dynamically adapts to grid conditions and phase balance status. Instead of aggressive static load shedding, the system continuously adjusts the rate and target of load reduction to maintain phase balance, resolving the contradiction between energy cost reduction and phase stability through dynamic control.
Data Source
AI summary
A method for power management of a computing system having two or more physical servers for hosting virtual machines of a virtual system and one or more uninterruptible power supplies for supplying at least a subset of the physical servers with power, each of the one or more uninterruptible power supplies being connected to a phase of a multiple phase power supply, is disclosed. The method comprises receiving an action input for the computing system, which may impact the power consumption of the physical servers, processing the received action input with a predictive model of power consumption of the physical servers with regard to the battery autonomy of the one or more uninterruptible power supplies and/or the load balancing of the several phases of the multiple phase power supply, and optimizing the utilization of the physical servers based on the result of the processing.