Active Queue Management for Input-Buffered Switches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The traditional Random Early Discard (RED) algorithm faces scalability issues and challenges in implementing Active Queue Management (AQM) for input-buffered network devices, as it requires multiple instances for thousands of Virtual Output Queues (VOQs, making it difficult and expensive to provide accurate queue length estimation and affecting uncongested flows.
Innovation Solution
A novel AQM method that sets a global drop flag based on the lesser of running and instantaneous buffer occupancy, with probabilities dependent on the number of active VOQs, allowing for adaptive AQM responses such as packet dropping or marking, and explicit congestion notifications, while protecting uncongested VOQs by adjusting thresholds based on traffic classes and buffer occupancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional RED algorithm is applied to each VOQ in input-buffered switches, then congestion control is improved, but device complexity and implementation cost increase significantly due to the need for thousands of queue length estimators
Solution Approach 1:
The patent merges the RED algorithm implementation from individual VOQ level to the input buffer level. Instead of implementing separate RED instances for each VOQ (which would require thousands of queue length estimators), the system uses a single RED instance that operates on the aggregate input buffer, thereby dramatically reducing device complexity while maintaining congestion control effectiveness.
Solution Approach 2:
The input buffer serves multiple functions: it acts as a shared buffer for multiple VOQs while also serving as the queue for the RED algorithm. This universal approach allows one RED instance to manage congestion for all VOQs simultaneously, eliminating the need for dedicated RED instances per VOQ and reducing implementation complexity.
2Device complexity
If traditional RED algorithm is applied to the entire input buffer, then device complexity is reduced, but uncongested VOQs are adversely affected due to lack of per-VOQ congestion awareness
Solution Approach 1:
The patent introduces local quality by differentiating between congested and uncongested VOQs. The system maintains per-VOQ information about packet counts and applies different handling strategies: for congested VOQs, packets are dropped according to RED probabilities, while for uncongested VOQs, packets are preserved. This selective approach protects uncongested flows from the harmful effects of aggregate buffer management.
Solution Approach 2:
The patent segments the input buffer into multiple virtual output queues (VOQs) for management purposes. Although the RED algorithm operates on the aggregate buffer, the system tracks and manages packets at the VOQ level, allowing differentiated treatment of congested versus uncongested queues. This segmentation enables precise control over which packets are subject to RED dropping.
3Measurement precision
If per-VOQ AQM is implemented for thousands of VOQs, then congestion control precision is improved, but measurement and estimation costs increase significantly
Solution Approach 1:
The patent merges the measurement and estimation functions into a single aggregate level. Instead of maintaining separate queue length estimators for each VOQ (which would be thousands of estimators requiring significant computational resources), the system uses one estimator that measures the total input buffer occupancy. This dramatically reduces estimation costs while still providing sufficient precision for congestion control.
Data Source
AI summary
Novel methods and devices are provided for AQM of input-buffered network devices. Preferred implementations of the invention control overall buffer occupancy while protecting uncongested individual VOQs. The probability of setting a “global drop flag” (which is not necessarily used to trigger packet drops, but may also be used to trigger other AQM responses) may depend, at least in part, on the lesser of a running average of buffer occupancy and instantaneous buffer occupancy. In some preferred embodiments, this probability also depends on the number of active VOQs. Moreover, a global drop flag is set in conjunction with a drop threshold M associated with the VOQs. Whether an AQM response is made may depend on whether a global drop flag has been set and whether a destination VOQ contains M or more packets. Different M values may be established for different classes of traffic, e.g., with higher M values for higher-priority traffic. AQM responses (e.g., to drop packets) may be taken more aggressively when there is a larger number of active VOQs.