Data Center Power Forecasting for Hotspot-Aware Workload Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches lack the ability to dynamically predict power requirements of servers based on historical usage patterns, leading to inefficiencies and increased costs in data center operations.
Innovation Solution
Utilizing machine learning models, particularly FBProphet, SARIMA/SARIMAX, Holt-Winter-ES, and Gated Recurrent Unit (GRU) Networks, to forecast power consumption and identify hotspots and coldspots in data centers, enabling efficient workload scheduling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If standard input power supply generation strategy is used, then infrastructure planning is simplified, but cost increases and performance and health of data center deteriorates
Solution Approach 1:
The patent implements dynamic power supply strategies by using machine learning models to forecast power consumption and identify hotspots and coldspots in real-time. This allows the power supply system to adapt dynamically to changing workload patterns, allocating power efficiently to servers that need it while reducing power to servers that don't, thereby improving overall power supply efficiency without complicating infrastructure planning
Solution Approach 2:
The system incorporates feedback mechanisms by continuously monitoring actual power consumption, comparing it with forecasted values, and using this information to adjust power allocation decisions. The identification of hotspots and coldspots provides feedback that drives dynamic power supply adjustments, creating a closed-loop system that optimizes energy efficiency while maintaining operational simplicity
2Ease of manufacture
If standard input power supply generation strategy is used, then infrastructure planning is simplified, but cost increases
Solution Approach 1:
The dynamic power supply approach uses ML-based forecasting to identify coldspots (servers with low power consumption) and reduces power allocation during these periods. This dynamic adjustment directly reduces operational costs by avoiding unnecessary power consumption while maintaining the simplicity of standard infrastructure planning through automated decision-making
Solution Approach 2:
The system changes the parameter of power allocation dynamically based on forecasted power consumption patterns. By adjusting power supply parameters according to predicted hotspots and coldspots, the system reduces operational costs without requiring changes to the physical infrastructure or planning processes
3Reliability
If power supply is increased to prevent overloading, then server health is maintained, but power consumption efficiency deteriorates
Solution Approach 1:
The system dynamically adjusts power supply levels based on real-time and forecasted server workload patterns. By identifying hotspots (servers approaching power limits) and coldspots (servers with excess power capacity), the system maintains server health by preventing overloading of critical servers while reducing power to non-critical servers, thereby maintaining reliability without sacrificing energy efficiency
Solution Approach 2:
The patent applies different power supply strategies to different servers based on their individual characteristics and workload patterns. Instead of a uniform power supply approach, the system identifies which specific servers need power protection (hotspots) and which can tolerate reduced power (coldspots), applying local quality adjustments that maintain overall system reliability while improving energy efficiency
Data Source
AI summary
Systems and methods are provided for using historic input power periodic data from a server in an IT data center to train a machine learning (ML) model to obtain forecasted power consumption data of the server for a future time period. Time windows of hotspots or coldspots are then identified in the forecasted power consumption data, hotspots being defined as areas or regions of over-utilization in a time series data, and coldspots being defined as areas or regions of under-utilization in a time series data. The hotspots and coldspots are identified by calculating an exponential mean average (EMA) of the forecasted power consumption data, taking points above the EMA as hotspots and points below the EMA as coldspots. The identified hotspots and coldspots can be used to schedule workloads for a server or a data center, to more efficiently plan existing workloads, or to introduce new workloads at more optimal time periods.