Predictive Bandwidth Models for Network Path Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optimizing network performance for specific applications in enterprise networks is challenging due to the use of the same protocols by business and non-business critical applications, leading to difficulties in distinguishing and optimizing traffic flows, and traditional reactive techniques result in periods of reduced performance before corrective measures are taken.
Innovation Solution
A predictive performance architecture that uses machine learning techniques to generate models for predicting available bandwidth and network performance, allowing for proactive adjustments to ensure service level agreements (SLAs) are met, by sending probing traffic along network paths and updating models based on confidence scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If reactive techniques are used to enforce network performance criteria, then network adjustments can be made based on monitored performance, but the network experiences periods of reduced performance before corrective measures are taken
Solution Approach 1:
The patent implements predictive modeling that analyzes historical network data and traffic patterns to forecast future performance conditions. The system proactively identifies potential SLA violations before they occur and executes preventive actions, such as adjusting QoS parameters or reallocating bandwidth, to ensure performance criteria are met without waiting for degradation to manifest.
2Device complexity
If traditional reactive techniques are used, then network engineering can be simplified with basic monitoring and adjustment, but the network cannot distinguish between business and non-business critical traffic flows using the same protocols
Solution Approach 1:
The patent introduces an intermediary predictive modeling layer that sits between network monitoring and traffic management functions. This model processes multiple data sources including protocol information, traffic patterns, and historical performance data to infer application identity and criticality. The model outputs enriched traffic classification results that enable differentiated QoS treatment without requiring complex signature-based detection mechanisms.
3Reliability
If proactive predictive modeling is implemented, then future network requirements can be predicted and corrective measures taken in advance, but additional data collection and model updating processes are required
Solution Approach 1:
The patent implements a feedback mechanism where actual network performance measurements are continuously compared against predicted values. When deviations exceed predefined thresholds, the system triggers model retraining using the new data, automatically adjusting prediction accuracy over time. This closed-loop approach ensures the model adapts to changing network conditions while maintaining high SLA compliance predictions.
Data Source
AI summary
In one embodiment, a network device receives metrics regarding a path in the network. A predictive model is generated using the received metrics and is operable to predict available bandwidth along the path for a particular type of traffic. A determination is made as to whether a confidence score for the predictive model is below a confidence threshold associated with the particular type of traffic. The device obtains additional data regarding the path based on a determination that the confidence score is below the confidence threshold. The predictive model is updated using the additional data regarding the path.