Backward Congestion Notification Rate Limiting for Data Center Ethernet
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing congestion management techniques, such as RED and ECN, are ineffective in high-speed, low-latency networks like Data Center Ethernet due to scalability issues and rapid buffer filling, leading to unnecessary slowdowns and throughput loss.
Innovation Solution
Implementing a congestion management method that uses backward congestion notification (BCN) messages to identify and mitigate congestion by rate-limiting traffic flows at reaction points, with parameters tuned based on feedback from congestion points, allowing for dynamic adjustment of traffic rates to prevent buffer overflow.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If RED or ECN techniques are used for congestion management, then congestion control is provided, but buffers fill up instantly causing excessive slowdowns and throughput loss in high-speed low-latency networks
Solution Approach 1:
The patent applies preliminary action by rate-limiting traffic flows at reaction points (ingress ports) before congestion occurs at the congestion point. BCN messages trigger rate limiters to proactively reduce traffic rates, preventing buffer overflow rather than reacting after buffers fill up. This is evident in the feedback loop where BCN messages cause rate limiters to adjust rates before congestion fully develops.
Solution Approach 2:
The patent implements feedback through BCN (Backward Congestion Notification) messages that travel from congestion points back to reaction points. These messages carry congestion information that triggers rate limiter adjustments. The system continuously monitors queue lengths and sends BCN messages when thresholds are exceeded, creating a closed-loop feedback control system that dynamically adjusts traffic rates based on actual congestion conditions.
2Quantity of substance
If buffer sizes are increased to handle high-speed traffic, then congestion management capacity is improved, but latency increases and the fundamental problem of rapid buffer filling persists
Solution Approach 1:
The patent introduces rate limiters as intermediary devices between traffic sources and network buffers. These rate limiters act as mediators that control the flow of traffic into the network, preventing buffers from filling up rapidly in the first place. By placing rate limiters at ingress ports (reaction points), the system controls traffic admission rather than relying on larger buffers to absorb traffic surges.
Solution Approach 2:
The system performs preliminary rate limiting at reaction points before traffic reaches congestion points. BCN messages trigger rate limiters to proactively reduce traffic rates, preventing the need for large buffers. This preliminary control action occurs upstream, stopping excessive traffic before it can fill buffers, thereby maintaining low latency while managing congestion capacity.
3Reliability
If link-level flow control is used to guarantee zero packet loss, then reliability is improved, but congestion spreads instantly throughout the network
Solution Approach 1:
The patent extracts the congestion control function from the link layer and implements it at the network layer through BCN messages and rate limiters. Instead of using link-level flow control that propagates congestion instantly across the entire network, the system extracts congestion management to specific reaction points (ingress ports) that receive BCN messages. This localized control prevents widespread congestion propagation while maintaining zero packet loss through targeted rate limiting.
Data Source
AI summary
The present invention provides improved methods and devices for managing network congestion. Preferred implementations of the invention allow congestion to be pushed from congestion points in the core of a network to reaction points, which may be edge devices, host devices or components thereof. Preferably, rate limiters shape individual flows of the reaction points that are causing congestion. Parameters of these rate limiters are preferably tuned based on feedback from congestion points, e.g., in the form of backward congestion notification (“BCN”) messages. In some implementations, such BCN messages include congestion change information and at least one instantaneous measure of congestion. The instantaneous measure(s) of congestion may be relative to a threshold of a particular queue and/or relative to a threshold of a buffer that includes a plurality of queues.