Network Bandwidth Estimation via Component Probability Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating bandwidth in computer networks are limited, as they require previous observations of the exact same network path and cannot estimate individual component bandwidths without visibility of these components, making it difficult to determine bottlenecks or provide accurate bandwidth estimates for new paths.
Innovation Solution
A method that collects end-to-end bandwidth observations from past communications across different paths to model and estimate individual network component bandwidths using probability functions, allowing for the estimation of component bandwidths without direct observation, and predicts end-to-end bandwidth for new paths based on these component estimates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing bandwidth estimation methods are used, then bandwidth can be estimated for paths with previous observations, but accurate bandwidth estimation cannot be obtained for new paths without previous measurements
Solution Approach 1:
The patent segments the network path into individual network components (routers, servers, relays). Instead of treating the path as a whole, it models each component's bandwidth separately using probability density functions. This segmentation allows the system to estimate bandwidth for new paths by combining component-level models, rather than requiring end-to-end path observations.
Solution Approach 2:
The patent performs preliminary action by building component-level bandwidth models from historical observations before they are needed for new path estimation. By pre-establishing probability density functions for each network component based on past data, the system prepares reusable models that can quickly estimate bandwidth for any new path without requiring new measurements.
2Loss of information
If traditional bandwidth estimation methods are used, then path bandwidth can be estimated, but individual component bandwidths cannot be determined without component visibility
Solution Approach 1:
The patent introduces an intermediary statistical model (probability density function) that bridges the gap between end-to-end path observations and individual component characteristics. By using PDFs to represent component bandwidth distributions, the system can infer component-level information from aggregate path measurements without requiring direct component visibility or additional measurement infrastructure.
Solution Approach 2:
The patent transforms the problem by changing from direct bandwidth measurement to probability density function parameter estimation. Instead of measuring component bandwidth directly, it estimates parameters (mean, variance) of probability distributions that characterize component behavior. This parameter transformation enables component-level insights from path-level observations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method comprising: collecting respective observations of end-end bandwidth experienced on different occasions by multiple past communications occurring over different respective observed paths over a network, each path comprising a respective plurality of network components; modelling each of the respective network components with a bandwidth probability function characterized by one or more parameters; and estimating a component bandwidth or component bandwidth probability density for each of the network components based on the modelling, by determining respective values for said parameters such that a combination of the component bandwidths or bandwidth probability densities for the network components in the observed paths approximately matches, according to an optimization process, the observations of the end-to-end bandwidth experienced by the past communications over the observed paths.