Packet Switch Congestion Control via Multi-Level Flow Regulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing packet switch technologies face challenges in achieving efficient and accurate congestion control, particularly in preventing packet discards and fully utilizing switch fabric capacity while supporting quality of service (QoS) and maintaining simplicity, especially with off-the-shelf packet processors.
Innovation Solution
A method that differentiates congestion levels for high and low priority traffic, using incremental approaches to adjust transmission rates based on congestion severity, with Flow Control commands (FC CMDs) such as 'START', 'SLOW', 'SLOW SHARP', and 'PAUSE' to regulate traffic flows, and utilizes source and fabric flow control mechanisms to manage buffer and queue conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional congestion control mechanisms are used, then packet discards occur during congestion, but implementing sophisticated control mechanisms increases device complexity
Solution Approach 1:
The patent segments congestion control into multiple discrete levels (Level 0-3) with specific buffer threshold ranges for each level. This segmentation allows the system to apply appropriate control intensity for each congestion severity, preventing packet discards through structured threshold-based control without requiring overly complex continuous control mechanisms.
Solution Approach 2:
The patent implements feedback mechanisms where the congestion control level is dynamically adjusted based on real-time buffer occupancy monitoring. The system continuously monitors buffer levels, determines the current congestion level, and adjusts traffic admission and forwarding accordingly, creating a closed-loop control system that prevents packet discards while maintaining manageable complexity.
2Reliability
If traffic is throttled to prevent congestion, then packet discards are reduced, but switch fabric capacity utilization decreases
Solution Approach 1:
The patent employs dynamic congestion control where the throttling intensity adapts to current congestion conditions. When buffer occupancy is low (Level 0-1), minimal throttling is applied and fabric capacity is fully utilized. When congestion increases (Level 2-3), throttling intensity increases progressively. This dynamic adjustment ensures lossless operation while maximizing fabric capacity utilization under varying traffic conditions.
Solution Approach 2:
The system changes key parameters (traffic admission rate, forwarding rate, burst size) based on the determined congestion level. At different levels, different parameter sets are applied: Level 0 allows full rate, Level 1 reduces excess rate, Level 2 further reduces committed rate, and Level 3 applies minimal throttling only. This parameter-based control prevents packet discards while optimizing fabric utilization across different congestion scenarios.
3Measurement precision
If multiple congestion levels are implemented, then traffic control precision improves, but control mechanism complexity increases
Solution Approach 1:
The patent divides the continuous buffer occupancy range into discrete segmented levels (Level 0: 0-25%, Level 1: 25-50%, Level 2: 50-75%, Level 3: 75-100%). Each level corresponds to specific control actions and parameter adjustments. This segmentation provides precise congestion differentiation while maintaining implementation simplicity through clear threshold-based decision logic.
Solution Approach 2:
Different control strategies are applied locally to different congestion levels. Level 0-1 use gentle throttling with excess rate reduction, Level 2 uses moderate throttling with committed rate reduction, and Level 3 uses aggressive throttling with minimal rate allowed. This local quality approach enables precise control adaptation to congestion severity without requiring a single complex control mechanism.
4Reliability
If strict QoS guarantees are provided, then service differentiation is improved, but bandwidth allocation flexibility decreases
Solution Approach 1:
The patent adjusts QoS parameters (CIR, PIR, CBS, PBS) dynamically based on congestion level. During normal conditions (Level 0-1), full QoS guarantees are provided with configured CIR and PIR rates. During moderate congestion (Level 2), excess rate is reduced while maintaining committed rate guarantees. During severe congestion (Level 3), bandwidth allocation is restricted to minimal necessary rates. This parameter adaptation maintains QoS reliability while providing bandwidth flexibility across different congestion scenarios.
Data Source
AI summary
A method is provided for managing congestion of traffic in a packet switch, which comprises the steps of: providing more than two different congestion levels at the switch for handling traffic associated with at least two different priorities, one being a low priority traffic and another being a high priority traffic; for each traffic flow having a specific destination and priority, determining a current congestion level that matches a congestion control level which is selected from among the more than two congestion control levels, and associating a different set of operations for handling each respective traffic flow, depending on its priority and its current congestion level, and for each traffic flow, performing operations that match the respective associated set of operations, and wherein a rate at which these operations are performed for each respective traffic flow increases when a higher congestion level occurs at that respective traffic flow.


