Software Power Profiling for Server Energy Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional power management systems in data centers face challenges in accurately estimating server power demand, leading to inefficiencies and increased costs due to their inability to handle smaller-scale electrical systems, variances in power consumption, and the need for physical sensors, which are costly and complex to deploy across thousands of servers.
Innovation Solution
A software-based system that monitors power consumption by installing daemons on each server to collect and transmit parameters to a Master Server, using mathematical models to estimate power and energy consumption without additional hardware, allowing for real-time monitoring and accurate determination of individual server power demand.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical power meters are installed in each server to measure power consumption, then measurement precision is improved, but device complexity and cost increase significantly
Solution Approach 1:
The patent creates a virtual copy of the power measurement system through software daemons that run on each server. Instead of physical meters, virtual power meters are implemented as software components that collect operational parameters and calculate power consumption using mathematical models, eliminating the need for expensive physical hardware while maintaining measurement capability
Solution Approach 2:
The patent replaces the mechanical/physical power metering system with a software-based computational system. Power consumption is determined through mathematical calculations using operational parameters (CPU usage, memory usage, disk I/O, network activity) collected by daemons, substituting physical measurement devices with software-based estimation
2Measurement precision
If physical power meters are deployed across thousands of servers, then power consumption monitoring is improved, but manufacturing precision and deployment cost worsen due to wiring and installation complexity
Solution Approach 1:
The patent deploys virtual copies of power monitoring functionality as software daemons on each server rather than installing physical meters. This software-based approach eliminates the need for complex wiring infrastructure and makes deployment scalable across thousands of servers without proportional increases in installation complexity
Solution Approach 2:
The software daemon performs multiple functions: collecting operational parameters, transmitting data to the master server, and enabling power consumption calculation. This multi-functional software component replaces multiple separate physical devices and simplifies the overall deployment architecture
3Device complexity
If conventional linear power models are used for power estimation, then device complexity is reduced, but measurement precision deteriorates with average errors of at least 5%
Solution Approach 1:
The patent transitions from simple linear power models to complex non-linear mathematical models that capture the true behavior of power consumption. These advanced models use multiple operational parameters and non-linear relationships to accurately represent how power consumption varies with system state, significantly improving estimation accuracy
Solution Approach 2:
The patent implements a feedback mechanism where the master server continuously receives operational parameters from daemons, compares actual power consumption measurements with model predictions, and automatically recalibrates the mathematical models when discrepancies are detected. This closed-loop feedback system maintains high accuracy over time
4Device complexity
If rack-level power measurement is performed using PDU meters, then device complexity is minimized, but measurement precision deteriorates to rack-level resolution only
Solution Approach 1:
The patent segments the power measurement system to provide both aggregate rack-level measurements and individual server-level measurements. Software daemons on each server collect and report operational parameters to a master server, which calculates individual server power consumption and aggregates it for rack-level totals, providing multi-granularity measurement capability
Data Source
AI summary
Systems and methods for estimating power consumption in a network of computing devices are described. Operational information of a target server is periodically received and compared to benchmark data of a model of the target server. The operational information comprises performance data of the target server during a predefined time interval. Power consumption of the target server is estimated using the performance and benchmark data. The benchmark data is recalibrated if an error in the estimated power consumption is detected. An agent installed on the target server for collecting performance data is described. The target server can be a virtualized server, in which case, the agent acquires at least some of the performance data from a hypervisor of a physical server that hosts the target server.