FastSATME Traffic Matrix Estimation via Simulated Annealing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating source-to-destination traffic matrices in communication networks are inefficient, particularly for large networks, as they rely on unrealistic assumptions and are too slow for real-time traffic engineering, making it challenging to accurately determine traffic patterns and adapt to changing conditions.
Innovation Solution
The FastSATME method uses a simulated annealing algorithm to estimate traffic matrices as a probability distribution, allowing users to specify prior information and constraints, and performs incremental estimates based on previous steps to improve speed and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods (packet monitoring, gravity model, tomographic model) are used to estimate traffic matrices, then traffic matrix estimation can be performed, but the computation time is too long for real-time traffic engineering
Solution Approach 1:
The patent performs preliminary actions by pre-computing routing matrices and preparing network topology data before actual traffic estimation is needed. The routing matrix R is pre-calculated based on network topology and routing policies, allowing the simulated annealing algorithm to directly use this pre-prepared data during estimation, significantly reducing computation time while maintaining accuracy.
Solution Approach 2:
The patent changes the approach from exact mathematical solutions to probabilistic estimation using simulated annealing. Instead of seeking exact traffic matrix values that satisfy all constraints perfectly, the method uses temperature parameter T in the annealing process to balance between exploration of solution space and convergence to optimal solutions, enabling faster computation with acceptable accuracy.
2Quantity of substance
If packet-based monitoring is used to record source and destination addresses, then traffic statistics can be collected, but huge tables are required and the method is not feasible for large networks
Solution Approach 1:
The patent extracts only the necessary information from packet monitoring - specifically the source and destination addresses - and aggregates this data into compact routing matrices rather than storing complete packet records. The routing matrix R condenses the raw packet data into a structured format that captures traffic patterns without requiring huge tables, reducing data complexity while preserving essential traffic statistics.
Solution Approach 2:
The routing matrix R serves multiple functions: it represents network topology, encodes routing policies, and provides the basis for traffic estimation. This multi-functional data structure eliminates the need for separate data structures for each function, reducing overall system complexity while maintaining comprehensive traffic monitoring capabilities.
3Loss of information
If link counts are used to derive traffic matrices, then traffic information can be obtained, but mathematical certainty is lost and probabilistic inference is required
Solution Approach 1:
The patent implements feedback mechanisms through the simulated annealing algorithm that iteratively refines traffic matrix estimates based on observed link counts. The algorithm uses the fitness function to evaluate how well a candidate traffic matrix explains the observed link traffic, and adjusts the estimate through random perturbations and acceptance/rejection decisions, converging to a probabilistically accurate solution that maximizes information from limited observations.
Data Source
AI summary
The FastSATME method and system estimate source-to-destination traffic matrices using a simulated annealing algorithm, the traffic matrix estimation being represented as a probability distribution over the set of all possible matrices that satisfy a set of given constraints. The constraints explicitly encode information that the user knows about the network traffic as components of an objective function (a fitness function), that is then minimized using simulated annealing. With the method according to the invention, arbitrary constraints of any form can be included. FastSATME works over a series of time steps. At the first time step FastSATME acts the same as SATME but in subsequent time steps, the estimate of the traffic matrix at time t is based on the estimate at t-1.