Neural Network Resource Allocation for Dynamic Workloads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing dynamic resource allocation systems in computing fail to accurately account for varying resource requirements across different tasks and applications, particularly in scenarios where data privacy and security are concerns, as they often rely on visible data and third-party management.
Innovation Solution
A method and system utilizing a neural network model generated by monitoring application instances, collecting instance-level data with dedicated agents, processing it through local and global models to determine workload-based resource requirements, ensuring secure and task-specific resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing dynamic resource allocation systems are used, then resource allocation flexibility is improved, but measurement precision of resource requirements deteriorates because they rely on coarse factors like number of applications and tasks rather than actual data processing needs
Solution Approach 1:
The patent segments resource monitoring and allocation into multiple hierarchical levels: individual application instances are monitored separately by dedicated agents, each generates its own local model, then these are aggregated into global models per application, and finally into a system-wide neural network. This segmentation enables precise measurement of resource requirements at each level while maintaining overall system adaptability.
Solution Approach 2:
The patent replaces traditional mechanical/census-based resource allocation methods (counting applications and tasks) with a neural network-based intelligent system. The neural network model processes instance-level data from multiple applications to accurately predict resource requirements, substituting crude mechanical counting with sophisticated computational intelligence for precise measurement.
2Ease of operation
If third-party service providers manage applications, then ease of operation is improved, but data security deteriorates because personal and sensitive data must be exposed to external parties
Solution Approach 1:
The patent segments data processing and model generation into isolated units where each application instance is monitored by its own dedicated agent that collects only necessary instance-level data. This segmentation minimizes data exposure to the minimum required for resource allocation purposes, maintaining security while enabling third-party management.
Solution Approach 2:
The patent implements local quality by generating separate local models for each application instance based on its specific instance-level data, rather than using a single global model for all applications. This allows each application to maintain its own data characteristics and security boundaries while still contributing to overall system resource allocation through the neural network.
3Device complexity
If static resource allocation is used, then device complexity is reduced, but adaptability to changing resource requirements deteriorates
Solution Approach 1:
The patent implements dynamic resource allocation through a neural network model that continuously learns from instance-level data and adapts to changing resource requirements. The system dynamically generates local models for each application instance, aggregates them into global models, and updates the overall neural network model to reflect current system conditions, enabling automatic adaptation without manual reconfiguration.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
State of the art systems being used to perform resource allocation fail to consider dynamic resource requirements.' The disclosure herein generally relates to resource allocation in computing systems, and, more particularly, to a method and system for automated resource allocation in computing systems based on dynamically determined resource requirements. The system collects application instance level data by monitoring each application in the computing device being monitored, and generated a neural network model of the computing device by using the collected application instance level data. Working of the computing device is emulated using the neural network model to determine workload, and in turn resource requirements of the computing device, and accordingly one or more resource allocation recommendations are generated.