Processor Utilization Simulation for Data Center Power Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data center architectures face inefficiencies in processor utilization prediction and power management due to the complexity of CISC and RISC-based processors, leading to increased power consumption and costs.
Innovation Solution
A computer-implemented method using a machine learning processor utilization model trained with system log data and code feature data to predict and simulate processor utilization, enabling actions to optimize processor usage and reduce costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If CISC-based processors are used in data centers, then processing capability is improved, but power consumption increases
Solution Approach 1:
The patent applies parameter changes by training machine learning models to predict processor utilization patterns and using this information to dynamically adjust system configuration, including selecting between CISC and RISC processor types based on predicted workload characteristics. This allows optimization of the processing capability to power consumption ratio by changing the operational parameters of the data center infrastructure.
Solution Approach 2:
The patent implements preliminary action by using machine learning models to predict future processor utilization patterns before workloads are executed. These predictions enable advance planning of resource allocation, processor selection, and scheduling decisions, allowing the system to prepare optimal configurations in advance rather than reacting to actual workload demands.
2Power
If complex processor architectures are used, then processing power is improved, but system complexity increases
Solution Approach 1:
The patent introduces machine learning models as intermediary components that mediate between the complex processor architectures and the workload management system. These models predict processor utilization patterns, enabling the system to make informed decisions about resource allocation and processor selection without directly managing the complexity of the underlying CISC or RISC architectures.
Solution Approach 2:
The patent creates simplified representations (copies) of complex processor utilization patterns through machine learning models. Instead of directly managing the complexity of CISC or RISC processor architectures, the system uses predicted utilization patterns as proxies to make scheduling and allocation decisions, effectively copying the essential behavioral characteristics without the full complexity.
3Measurement precision
If processor utilization prediction accuracy is improved, then resource allocation is optimized, but computational overhead increases
Solution Approach 1:
The patent applies partial action by training machine learning models on historical processor utilization data to capture key patterns without attempting to model every possible variable. The models focus on the most significant predictors of processor utilization, achieving sufficient accuracy for resource allocation decisions while avoiding the computational overhead of exhaustive analysis.
Data Source
AI summary
A machine learning-based processor utilization prediction process is provided which includes training a processor utilization model using system log data, code feature data, and processor-associated data of a system to, at least in part, predict processor utilization to execute application code on the system. In addition, the process includes generating, using the processor utilization model, a processor utilization simulation for the system to execute the application code, and initiating an action based on the processor utilization simulation for the system to execute the application code.