Lightweight Regression Models for Network Device Resource Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing network device models consume significant computing resources and are difficult to update quickly, making them unsuitable for deployment in resource-constrained devices and prone to misconfiguration that can lead to faulty operation.

Innovation Solution

The development of lightweight models using regression analysis to predict operating states, allowing for efficient execution on devices with limited resources and preventing misconfiguration by predicting resource utilization before committing configuration changes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning models are used to predict resource utilization, then prediction accuracy is improved, but computing resource consumption increases significantly

Engineering Contradiction:

Improveprediction accuracyVSAvoidcomputing resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent replaces expensive, resource-intensive machine learning models with inexpensive regression analysis models that consume minimal computing resources. The regression models are lightweight and can be executed efficiently on resource-constrained network devices, achieving acceptable prediction accuracy without the heavy computational overhead of machine learning approaches.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

Solution Approach 2:

The patent changes the mathematical approach from complex machine learning algorithms to simpler regression analysis with configurable parameters. By adjusting regression parameters and using polynomial fitting, the system achieves a balance between prediction accuracy and computational efficiency, allowing the models to run on devices with limited resources.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If detailed data collection and extensive modeling are performed, then model accuracy is improved, but model update difficulty increases

Engineering Contradiction:

Improvemodel accuracyVSAvoidmodel update difficulty

Core Design Contradiction:

Measurement precisionVSEase of manufacture

Solution Approach 1:

The patent implements dynamic model updating capabilities where regression models can be quickly retrained and updated as devices change due to hardware upgrades or software updates. The system continuously collects operational data and updates the regression parameters accordingly, allowing models to adapt to changing device states without requiring extensive re-modeling efforts.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent performs preliminary data collection during test bed simulations to establish regression relationships before deployment. By pre-collecting configuration parameters and operational state data during testing, the system prepares regression models in advance that can be quickly deployed and updated, reducing the complexity of model updates when devices change.

Inventive Principle:

Principle #10Preliminary action

3Adaptability or versatility

If complex machine learning models are deployed, then prediction capability is improved, but device compatibility decreases

Engineering Contradiction:

Improveprediction capabilityVSAvoiddevice compatibility

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent replaces complex machine learning models with simple regression analysis models that can be executed on resource-constrained devices such as routers, switches, and other network equipment with limited processing power, memory, and energy resources. The regression models achieve acceptable prediction capability without requiring the computational infrastructure needed for machine learning.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

Solution Approach 2:

The patent substitutes the computational mechanism of machine learning with the mathematical mechanism of regression analysis. This substitution replaces resource-intensive iterative optimization algorithms with closed-form or iterative least-squares solutions that are computationally efficient and compatible with embedded network devices.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11797408B2Dynamic prediction of system resource requirement of network software in a live network using data driven models

Publication Date: 2023.10.24 JUNIPER NETWORKS INC

AI summary

In general, a device comprising a processor and a memory may be configured to perform various aspects of the techniques described in this disclosure. The processor may conduct, based on configuration parameters, each of a plurality of simulation iterations within the test environment to collect a corresponding plurality of simulation datasets representative of operating states of the network device. The processor may perform a regression analysis with respect to each of the plurality of configuration parameters and each of the plurality of simulation datasets to generate a light weight model representative of the network device that predicts an operating state of the network device. The processor may output the light weight model for use in a computing resource restricted network device to enable prediction of the operating state of the computing resource restricted network device when configured with the configuration parameters. The memory may store the light weight model.