Switch Queue Scheduling for Fair TCP and UDP Bandwidth

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing network switches face issues with unfair bandwidth allocation among data flows of different protocols, such as TCP and UDP, leading to congestion and inefficient resource utilization due to the use of first-in-first-out queues.

Innovation Solution

Implement a scheduling method that sorts data flows by sending rate, divides them into multiple queues based on fair share calculations, and assigns them to specific queues to ensure fair bandwidth allocation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If FIFO queues are used for packet forwarding in switches, then device complexity is reduced and cost is lowered, but unfair bandwidth allocation occurs among data flows with different protocols and sending rates

Engineering Contradiction:
Improvequeue structureVSAvoidbandwidth allocation fairness
Core Design Contradiction:
Device complexityVSReliability

Solution Approach 1:

The patent segments data flows into different protocol types (TCP flows and UDP flows) and assigns them to separate queues. This segmentation prevents TCP flows from being blocked by UDP flows and vice versa, ensuring fair bandwidth allocation while maintaining simple FIFO queue structures within each protocol-specific queue.

Inventive Principle:
Principle #1Segmentation

2Productivity

If multiple UDP flows with different sending rates share the same queue, then queue utilization is improved, but bandwidth allocation becomes unfair with higher rate flows obtaining disproportionate bandwidth

Engineering Contradiction:
Improvequeue utilizationVSAvoidbandwidth allocation fairness
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent segments UDP flows into multiple priority queues based on their sending rates. High-rate UDP flows are placed in lower priority queues while low-rate UDP flows are placed in higher priority queues, ensuring that low-rate flows receive fair bandwidth allocation rather than being overwhelmed by high-rate flows in a shared queue.

Inventive Principle:
Principle #1Segmentation

3Adaptability or versatility

If multiple TCP flows share the same queue, then each TCP flow can control its own bandwidth usage, but excess bandwidth from low-rate flows is captured by high-rate flows rather than being evenly distributed

Engineering Contradiction:
Improvebandwidth control capabilityVSAvoidbandwidth distribution fairness
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent segments TCP flows into different priority queues based on their sending rates relative to fair share. TCP flows using less than their fair share are placed in higher priority queues to capture excess bandwidth, while TCP flows exceeding their fair share are placed in lower priority queues. This ensures equitable bandwidth distribution while maintaining TCP's adaptive control capability.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS20250254130A1Switch and scheduling method for switch
Publication Date: 2025.08.07 ACCTON TECHNOLOGY CORPORATION
  • US20250254130A1 patent drawing
  • US20250254130A1 patent drawing
  • US20250254130A1 patent drawing

AI summary

A scheduling method for a switch includes steps as follows. First type protocol flows are sorted according to a sending rate to obtain a plurality of sorted first type protocol flows; a difference of sending rate between each two adjacent of sorted first type protocol flows is calculated to obtain a plurality of differences; N−1 differences are selected from the differences, where each of the N−1 differences is greater than a largest one of remaining differences; a position of each of the N−1 differences between corresponding two adjacent of sorted first type protocol flows is marked to obtain N−1 marked positions; the sorted first type protocol flows are divided into N groups of first type protocol flows according to the N−1 marked positions; each group of the N groups of first type protocol flows is directed to a first queue of the N first queues correspondingly.