Under-Determined Inverse Problem Solver for Network Flow Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is no efficient method to solve under-determined inverse problems in computer networks, which prevents the optimization of flows in such systems.
Innovation Solution
A method is developed to solve under-determined inverse problems by establishing a computer network with unknown per-flow size, delay, and throughput inferences, involving a learning phase, adaptive measurement and inference phases, and the computation of an optimal observation matrix using SNMP link loads and traffic matrices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional measurement methods are used in under-determined systems, then the system structure remains simple, but the ability to solve inverse problems and optimize flows is lost
Solution Approach 1:
The patent implements a learning phase before the adaptive measurement phase, where the system pre-computes the optimal observation matrix and identifies informative flows in advance. This preliminary action enables the system to solve under-determined inverse problems effectively when actual measurements are needed, resolving the contradiction between simple structure and precise measurement.
Solution Approach 2:
The patent introduces an optimal observation matrix as an intermediary computational structure that transforms the under-determined measurement problem into a solvable form. This matrix acts as a mediator between the limited measurements available and the complete flow information desired, enabling accurate flow estimation without requiring complex hardware modifications.
2Loss of information
If all unknown network flows are continuously measured, then complete flow information is obtained, but the measurement cost and system complexity increase significantly
Solution Approach 1:
The patent measures only a partial set of most informative unknown network flows rather than all flows continuously. The optimal observation matrix identifies which subset of flows provides the most information for reconstructing complete traffic matrix, achieving information completeness with reduced measurement complexity.
Solution Approach 2:
The system uses the measured flow data and optimal observation matrix to automatically infer and reconstruct the complete traffic matrix without requiring direct measurement of all flows. The measurement system serves itself by using partial measurements combined with computational inference to obtain complete information.
3Measurement precision
If the observation matrix is updated frequently to adapt to changing network conditions, then the estimation accuracy is improved, but the computational overhead and time consumption increase
Solution Approach 1:
The patent updates the optimal observation matrix periodically based on detected significant changes in network conditions rather than continuously. The system monitors for substantial changes in traffic patterns and triggers matrix updates only when necessary, maintaining estimation accuracy while reducing computational overhead and time consumption.
Data Source
AI summary
A method for solving an under-determined inverse problem or network inference/tomography problem in per-flow size, delay, loss and throughput inference in a computer network, through a system is presented. The method includes the following steps, which are not necessarily in order. First, establishing the computer network having a plurality of nodes wherein the per-flow size, the delay, the loss and the throughput inference are unknown. An original observation or routing matrix determines how flows are appeared on the links and construct the measurements. Next, performing a learning phase to obtain an optimal observation matrix or pseudo-optimal observation matrix. After that, performing a computer controller adaptive measurement and inference phase to estimate the set of unknowns using the measurement quantities, and a function of one of the set consisting of: the optimal observation matrix, the original observation matrix, or both.