Fairness and Congestion Control Using Rate-Window Ratios
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network congestion signaling methods in computing networks result in additional bandwidth overhead and inefficiencies, particularly in high-performance applications like AI and HPC, due to the use of end-to-end notifications and suboptimal congestion feedback mechanisms.
Innovation Solution
Implementing congestion feedback with a weighted random early detection mechanism that varies signaling frequency based on congestion level, using probabilistic generation of BTS packets and entropy recycling for efficient path selection, along with RTT and ECN signals for congestion control, and optimizing packet delivery contexts to minimize overhead and improve network performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If end-to-end congestion notification messages are sent to computing nodes contributing to congestion events, then congestion control is achieved, but bandwidth overhead increases
Solution Approach 1:
The patent implements back-to-sender (BTS) congestion notification where only the specific computing node causing congestion receives targeted notifications rather than all nodes receiving end-to-end notifications. This localized approach reduces unnecessary bandwidth consumption while maintaining effective congestion control at the source of the problem.
Solution Approach 2:
The system uses probabilistic generation of BTS packets based on congestion level thresholds. When congestion exceeds thresholds, BTS packets are generated with probabilities proportional to the congestion level, sending only the necessary amount of congestion feedback information rather than continuous or excessive notifications, thus optimizing bandwidth usage.
2Productivity
If congestion notification frequency is increased to improve congestion control responsiveness, then congestion management effectiveness improves, but bandwidth overhead increases
Solution Approach 1:
The patent dynamically adjusts BTS packet generation frequency based on real-time congestion levels. The generation probability is proportional to the congestion level, creating a dynamic feedback mechanism that increases notification frequency during high congestion and reduces it during normal conditions, optimizing both responsiveness and bandwidth efficiency.
Solution Approach 2:
The system implements periodic congestion level measurements and threshold-based BTS packet generation. Instead of continuous notifications, congestion is monitored at intervals and BTS packets are generated periodically when thresholds are exceeded, reducing overall bandwidth overhead while maintaining effective congestion control timing.
3Productivity
If multiple entropy values are stored for path selection to improve network routing efficiency, then path selection quality improves, but storage overhead increases
Solution Approach 1:
The patent implements an entropy recycling mechanism where acknowledged good entropy values are stored in a limited buffer and reused for future path selection. When the buffer is full, older entropy values are discarded and new ones are added. This recycling approach maximizes the utilization of stored entropy values while maintaining a compact storage structure.
Solution Approach 2:
Instead of storing extensive path information, the system stores compact entropy values that serve as keys for path selection. These entropy copies are sufficient for routing decisions without requiring storage of complete path details, significantly reducing storage overhead while maintaining routing efficiency.
Data Source
AI summary
A first ratio of a sending rate limit to a full line rate for a link in the computing network is accessed. A second ratio of a sending window size to W_max for the link is accessed. W_max is the maximum allowed window size or the window size that utilizes an end-to-end path for the link. One or more of the first or second ratio is used to determine an amount to reduce the sending rate or window for the link in response to an indication of network congestion in the link.