Virtualized Application Power Budgeting for Data Centers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face challenges in managing power allocation efficiently due to varying power consumption rates of applications, leading to conflicts between total available power and application requirements, with existing methods like power capping being non-discriminatory and inefficient.
Innovation Solution
Implementing virtualized application power budgeting, which uses a hierarchy of controllers to dynamically manage and distribute power based on application priority, tier structure, and resource utilization, allowing for reallocation of resources and adjustment of virtual machine power consumption through DVFS and scheduling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If power capping is applied across the entire server, then power consumption is reduced, but application performance is degraded uniformly without considering priority
Solution Approach 1:
The patent implements differentiated power management by applying different power caps to different applications based on their priority levels. High-priority applications receive higher power caps while low-priority applications receive lower caps, allowing localized optimization rather than uniform power reduction across all applications.
Solution Approach 2:
The system dynamically adjusts power caps based on real-time conditions including application priority, current power consumption, and performance metrics. The power cap for each application is not static but adapts to changing system states, allowing the system to respond to varying workloads and priority requirements.
2Productivity
If power is allocated based on peak requirements, then all applications can run at peak performance, but total power consumption exceeds available power capacity
Solution Approach 1:
Instead of allocating power for all applications to run at peak simultaneously, the system allocates partial power to each application based on priority, ensuring that the sum of power allocations does not exceed available capacity. High-priority applications receive sufficient power to run near peak while low-priority applications receive reduced power allocation.
Solution Approach 2:
The system continuously monitors actual power consumption and performance metrics of applications, using this feedback to dynamically adjust power cap allocations. This closed-loop control ensures that power is allocated efficiently based on actual needs rather than static peak requirements.
3Ease of operation
If computing resources are proportioned without prior knowledge of applications, then resource allocation is simplified, but resource utilization efficiency decreases
Solution Approach 1:
The system allows applications to self-report their power requirements and performance characteristics, eliminating the need for pre-configuration of resource allocation parameters. Each application provides information about its needs, and the power management system uses this self-reported data to automatically determine appropriate power caps and allocations.
Data Source
AI summary
Virtualized application power budgeting can manage power budgeting for multiple applications in data centers. This power budgeting may be done in intelligent and/or dynamic ways and may be useful for updating power budgets, resolving conflicts in requests for power, and may improve the efficiency of the distribution of power to multiple applications.Virtualized application power budgeting can distinguish between priority applications and non-priority applications at a granular, virtual machine level and reduce the power consumption to only non-priority applications when there are power consumption conflicts. Virtualized application power budgeting may be able to determine the most efficient manner of providing power to each application in a data center. Further, virtualized application power budgeting may be able to distribute power according to application priority and other predetermined requirements and improve the efficiency of the power consumption by the devices in the data center.