Network Node Preemption Control Beat-Down Effect
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing flow control mechanisms in computer networks often preempt too many flows due to inefficient marking of packets, leading to network congestion and Quality of Service (QoS) degradation, particularly in scenarios with multiple congested network elements, resulting in the 'beat-down' problem where more flows are dropped than necessary.
Innovation Solution
A node determines an overload ratio and marks unmarked traffic based on a calculated need ratio, distinguishing between constrained and unconstrained inputs to ensure fair share usage of output bandwidth, thereby minimizing unnecessary preemption and maintaining network connectivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If packets are marked at multiple congested network elements to reflect congestion severity, then congestion detection accuracy is improved, but flow preemption becomes excessive due to cumulative marking effects
Solution Approach 1:
The patent applies preliminary anti-action by having network elements compensate for previously marked packets before making new marking decisions. When a packet is marked at an upstream element, downstream elements recognize this prior action and adjust their marking behavior accordingly, preventing cumulative over-marking. This anticipatory compensation mechanism counteracts the beat-down effect before it fully manifests.
Solution Approach 2:
The patent implements feedback mechanisms where network elements track and communicate marking status across multiple hops. Information about packets already marked at upstream elements is fed back to downstream elements, enabling them to make informed decisions about whether additional marking is necessary. This feedback loop prevents redundant marking and the resulting excessive preemption.
2Reliability
If per-packet processing is performed to mark packets exceeding preemption threshold, then QoS protection is improved, but network overhead and complexity increase
Solution Approach 1:
The patent segments the QoS protection function across multiple network elements along the data path. Instead of concentrating all marking decisions at a single point, each network element performs localized marking based on its own congestion observations and knowledge of upstream marking actions. This segmentation distributes processing complexity while maintaining comprehensive QoS protection.
Solution Approach 2:
The patent applies preliminary action by having network elements prepare and exchange information about congestion conditions and marking decisions in advance. Elements proactively communicate their marking status and congestion levels to neighboring elements, enabling coordinated QoS protection without requiring complex real-time analysis at each decision point.
3Stability of the object's composition
If aggressive rate reduction is applied to alleviate congestion, then network stability is improved, but link utilization decreases due to beat-down effect
Solution Approach 1:
The patent applies partial action by having each network element perform marking at a moderate level rather than aggressive rate reduction. By distributing the congestion mitigation effort across multiple elements, each element can apply milder marking that maintains link utilization while collectively achieving the necessary congestion control. This avoids the beat-down effect of excessive single-point marking.
Data Source
AI summary
In one embodiment, a node determines a ratio of traffic that is to be marked at an output. The node also determines if an input feeding the output has less traffic marked than is to be marked at the output, and marks unmarked traffic of that particular input corresponding to a ratio substantially equal to the difference between what is needed at the output and what is marked at the input. Also, in one embodiment, a node may determine a fair share of an output for unconstrained inputs (e.g., a preemption threshold divided among the inputs, taking into consideration inputs that use less than their fair share), and mark traffic accordingly.