Bandwidth Estimation via One-Way Delay CDF Knee Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing bandwidth estimation techniques fail to accurately estimate available bandwidth in non-FIFO scheduling and frame-level contention environments, particularly in broadband access networks like cable modems and 802.11-based wireless networks, due to assumptions of well-defined raw bandwidth, FIFO packet scheduling, and bursty cross-traffic patterns.
Innovation Solution
A method that estimates available bandwidth by probing for idle periods in the link by gathering one-way delay samples, normalizing them, and identifying the knee in the cumulative distribution function (CDF) to determine the fraction of time the channel is idle, then multiplying by the link capacity to obtain an estimate of available bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional bandwidth estimation techniques are used, then measurement simplicity is maintained, but measurement precision deteriorates in non-FIFO scheduling and frame-level contention environments
Solution Approach 1:
The patent segments the bandwidth estimation problem into distinct phases: probe packet transmission, idle period detection through OWD sampling, CDF construction, and knee point identification. Each phase handles a specific aspect of the measurement, allowing the system to accurately capture bandwidth characteristics in non-FIFO environments without requiring complete redesign of the measurement approach
Solution Approach 2:
The patent introduces dynamic adaptation by using the knee point detection algorithm that automatically adjusts to varying network conditions. The system dynamically identifies the transition point in the CDF where queued packets begin to dominate, allowing accurate bandwidth estimation even when traffic patterns and scheduling behaviors change over time
2Measurement precision
If probe packets are sent to sample one-way delay, then bandwidth estimation accuracy is improved, but network perturbation increases
Solution Approach 1:
The patent applies partial action by sending a limited number of probe packets (e.g., 200 packets over 5 seconds) rather than continuous probing. This sufficient but not excessive sampling provides accurate bandwidth estimation while minimizing the perturbation introduced to the network, balancing measurement needs with network stability
Solution Approach 2:
The patent implements periodic bandwidth estimation by conducting measurements at intervals rather than continuously. The system gathers OWD samples over a defined period (e.g., 5 seconds), constructs the CDF, identifies the knee point, and then can repeat the process later, reducing overall network perturbation while maintaining estimation accuracy
Data Source
AI summary
Measurement systems according to embodiments of the invention allow accurate bandwidth estimation even in non-FIFO scheduling and frame-level contention environments. In an embodiment of the invention, the approximate fraction of time that a link is idle is found by probing for idle periods (“gaps”). The fraction of idle time is then multiplied by the capacity to obtain an estimate of the available bandwidth. Gap time is estimated in an embodiment of the invention fraction by gathering samples of one-way delay (OWD) over the link in question. After the OWD is normalized and plotted, the knee in the cumulative distribution function (CDF) of OWD samples is used to identify the fraction of time that the channel is idle.