Precedence Drop QoS for Latency Differentiation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional QoS implementations, such as Per-flow QoS, require complex router state and signaling, leading to high equipment costs and management complexity, while simpler alternatives like DiffServ are more expensive but lack performance guarantees, and existing methods struggle with packet reordering and congestion control.
Innovation Solution
The Precedence Drop Quality of Service (PDQoS) methodology assigns a drop precedence value to each packet, allowing for congestion control through packet dropping without multiple queues or complex schedulers, replacing RED with simpler operations and providing user-controlled QoS enforcement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Per-flow QoS is implemented to provide performance guarantees, then QoS enforcement capability is improved, but device complexity and management complexity increase
Solution Approach 1:
The patent segments QoS handling into two parts: (1) Per-flow QoS marking at network edges where packets are classified and marked with drop precedence values, and (2) Simple Per-class QoS enforcement at core routers where packets are handled based on their markings without complex per-flow state. This segmentation allows performance guarantees while reducing core router complexity.
Solution Approach 2:
The patent extracts the complex per-flow QoS processing from core routers and relocates it to network edge devices. Core routers only need to maintain simple per-class state and enforce QoS based on packet markings, removing the burden of complex router state and signaling from the core network.
2Reliability
If multiple queues and complex schedulers are used for congestion control, then QoS differentiation is improved, but device complexity increases
Solution Approach 1:
The patent removes complex multi-queue scheduling mechanisms from core routers and replaces them with simple drop precedence-based congestion control. QoS differentiation is achieved through packet marking rather than through complex queue management, significantly reducing device complexity.
Solution Approach 2:
The patent changes the QoS differentiation approach from structural (multiple queues with different scheduling policies) to parametric (single queue with drop precedence parameters). Each packet carries a drop precedence value that determines its treatment during congestion, achieving QoS differentiation without multiple queues or complex schedulers.
3Reliability
If RED algorithm is used for congestion control, then congestion management is improved, but processing complexity increases
Solution Approach 1:
The patent simplifies congestion control by changing from RED's probabilistic drop based on average queue depth to a deterministic approach based on drop precedence values. During congestion, packets are dropped based on their marked precedence rather than through complex probabilistic calculations, reducing processing complexity while maintaining effective congestion control.
Data Source
AI summary
A packet is classified into a class. A priority value is assigned to the packet wherein packets in a flow are assigned priorities according to some probability distribution within some band. A determination is made, at a network device for a highest latency class, whether a sum of queued packet sizes of previously received packets having an equal or smaller latency class than the packet and larger or equal priority than the packet is larger than a threshold value. When the sum is larger, the packet is dropped, otherwise a determination is made whether a latency class of the packet is less than the latency class of the network device. When the latency class is not less, the packet is stored in a queue for the latency class. When the latency class is less, then the process is repeated until the packet is dropped or stored in a queue.