Proactive Link Load Balancing via Health Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In networks, it is challenging to identify links in good states, determine the cause of links entering bad states, predict future link states, or proactively prevent links from deteriorating, leading to suboptimal data transmission quality and increased latency, packet loss, and jitter.
Innovation Solution
A device, such as a network appliance or middlebox, uses predictive algorithms combining static rules and online machine learning classifiers to assess link health by monitoring parameters like packet loss, latency, and jitter, and takes proactive actions to maintain link quality by redistributing traffic and preventing overloading.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If load balancing is performed by selecting links in good state, then network throughput and reliability are improved, but the complexity of link state monitoring and prediction increases
Solution Approach 1:
The system performs preliminary actions by proactively predicting link health states before actual degradation occurs. The machine learning classifier analyzes current link parameters (packet loss, latency, jitter) to predict future states, enabling preventive load balancing decisions that maintain reliability without requiring complex real-time intervention
Solution Approach 2:
The link state monitoring system serves itself by using the same link parameters (packet loss, latency, jitter) for both quality assessment and predictive analysis. The system automatically updates its prediction models using observed link behavior, reducing the need for external complex monitoring infrastructure
2Reliability
If traffic is concentrated on links in good state, then transmission quality is improved, but the risk of cascading failures increases
Solution Approach 1:
The system applies preliminary anti-action by predicting which links are at risk of entering bad states and proactively redistributing traffic before failures occur. The machine learning classifier identifies links showing early signs of degradation and preemptively balances load away from them, preventing cascading failures while maintaining transmission quality on healthy links
Solution Approach 2:
The system continuously monitors link parameters (packet loss, latency, jitter) and uses this feedback to dynamically adjust load balancing decisions. The feedback loop enables the system to detect early signs of link degradation and redistribute traffic before failures cascade, maintaining transmission quality while preventing system-wide issues
3Reliability
If proactive link health prediction is implemented, then preventive load balancing is enabled, but the computational resources required increase
Solution Approach 1:
The system applies partial action by using simplified prediction models that focus on the most critical link parameters (packet loss, latency, jitter) rather than analyzing all possible link characteristics. This approach provides sufficient prediction accuracy for load balancing decisions while consuming fewer computational resources compared to comprehensive analysis methods
Solution Approach 2:
The system changes parameters by using lightweight machine learning models that process link state data in efficient formats. The prediction algorithms are optimized to work with standardized network metrics, reducing computational overhead while maintaining prediction accuracy sufficient for effective load balancing
Data Source
AI summary
Disclosed embodiments describe systems and methods for predicting health of a link. A device in communication with a link can identify profile information of a stream of network traffic traversing the link. The device can determine a first prediction of health of the link by applying one or more rules to the plurality of parameters of the profile information. The device can determine a second prediction of health of the link by applying a classifier to one or more timed sequences of the plurality of parameters of the profile information. The device can establishes a respective weight for each of the first prediction of health and the second prediction of heath. The device can select, using the respective weight, between the first prediction of health and the second prediction of health to provide a predictor of the health of the link.