Network Distance Prediction via Low-Rank Sparse Factorization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network coordinate systems face inaccuracies in predicting network distances due to random network delay fluctuations, machine faults, and measurement errors, leading to large network distance errors during prediction.
Innovation Solution
A method involving a to-be-positioned node communicating with reference nodes to construct a local distance matrix, performing low-rank sparse factorization to remove error and outlier values, and using the resulting low-rank matrix to determine more accurate network distances and coordinates, thereby improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If network distance measurement is performed using traditional methods, then measurement overhead is reduced, but prediction accuracy deteriorates due to error and outlier values
Solution Approach 1:
The patent extracts and removes error values and outlier values from the distance measurement data through low-rank sparse factorization. This separates the reliable distance information from the corrupted measurements, thereby improving prediction accuracy while maintaining measurement reliability.
Solution Approach 2:
The patent introduces low-rank sparse factorization as an intermediary processing step between raw distance measurement and coordinate fitting. This intermediary mechanism filters out erroneous data while preserving valid measurements, resolving the contradiction between accuracy and reliability.
2Measurement precision
If coordinates are fitted using measured network distances, then prediction functionality is achieved, but accuracy deteriorates due to random network delay pollution and measurement errors
Solution Approach 1:
The patent converts the harmful effect of random network delay pollution into a benefit by using low-rank sparse factorization to identify and remove the pollution. The factorization process distinguishes between systematic patterns (beneficial) and random noise (harmful), thereby improving coordinate fitting accuracy.
Solution Approach 2:
The patent changes the parameter representation by transforming raw distance measurements into a low-rank matrix structure. This parameter transformation enables the separation of signal from noise, improving the accuracy of coordinate fitting despite the presence of network delay pollution.
3Loss of information
If all nodes perform end-to-end measurement, then measurement completeness is improved, but measurement complexity increases significantly
Solution Approach 1:
The patent makes the network coordinate system universally applicable by enabling accurate distance prediction between any pair of nodes through coordinate fitting. Once coordinates are determined, the system can predict distances for all node pairs without requiring actual measurements, achieving measurement completeness with minimal complexity.
Solution Approach 2:
The patent creates a virtual copy of the network topology in the form of a coordinate system. Instead of performing physical measurements between all node pairs, the system uses the coordinate model to calculate distances, thereby achieving complete measurement information with O(M) complexity rather than O(M²).
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
Embodiments of the present invention provide a network distance prediction method and apparatus. The method includes: communicating, by a to-be-positioned node, with at least two reference nodes, to determine values of at least some elements in a local distance matrix; constructing, by the to-be-positioned node, the local distance matrix based on the values of the at least some elements in the local distance matrix; performing, by the to-be-positioned node, low-rank sparse factorization on the local distance matrix, to obtain a low-rank matrix; obtaining, by the to-be-positioned node, values of elements in a first element set of the low-rank matrix, to use the values as target values of network distances between the to-be-positioned node and the at least two reference nodes; communicating, by the to-be-positioned node, with the reference nodes, to obtain coordinates of the reference nodes in a network coordinate system; and determining, by the to-be-positioned node, coordinates of the to-be-positioned node based on the target values of the network distances between the to-be-positioned node and the reference nodes and the coordinates of the reference nodes in the network coordinate system. The embodiments of the present invention can improve accuracy of network distance prediction.