Lossless Ethernet Congestion Isolation via Random Hashing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current data center networks face issues with network overload and packet loss due to collisions exceeding buffer capacity, particularly in large-scale, low-latency environments, where priority-based flow control leads to head-of-line blocking, affecting well-behaved flows and increasing network complexity.

Innovation Solution

Implementing a cost-effective congestion isolation method by designating congested priority queues and using randomization to isolate data flows, with downstream devices computing hash buckets for data packets and sending congestion isolation messages to upstream devices, allowing for efficient queuing and minimizing HOL blocking without requiring per-flow queuing or packet discard.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If priority-based flow control is implemented to prevent packet loss, then reliability is improved, but head-of-line blocking occurs affecting well-behaved flows

Engineering Contradiction:

Improvepacket loss preventionVSAvoidnetwork throughput

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent segments the network buffer into multiple priority queues (different service classes) and further segments each queue into isolated and non-isolated portions. This segmentation allows the system to isolate only the congested flows into specific queues while allowing well-behaved flows to access non-isolated portions, preventing HOL blocking from affecting all flows uniformly.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by treating different flows differently based on their congestion status. Congested flows are directed to isolated queues while well-behaved flows access non-isolated queues. This localized differentiation ensures that flow isolation is applied only where necessary (to congested flows) rather than globally to all flows, maintaining productivity for well-behaved flows.

Inventive Principle:

Principle #3Local quality

2Productivity

If per-flow queuing is implemented to isolate congested flows, then network efficiency is improved, but device complexity increases

Engineering Contradiction:

Improvenetwork efficiencyVSAvoidqueuing structure complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent merges the concepts of priority-based flow control and flow isolation into a unified queuing structure. Instead of implementing separate per-flow queues for each flow (which would be complex), it combines multiple service classes with isolated and non-isolated portions within each class, reducing the number of queues needed while maintaining isolation capabilities.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The patent makes each priority queue multi-functional by providing both isolated and non-isolated portions within the same queue structure. This allows a single queue to serve multiple purposes: handling both congested and non-congested flows for the same service class, eliminating the need for separate queue structures and reducing overall device complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Productivity

If congestion isolation messages are sent upstream to reduce HOL blocking, then network efficiency is improved, but network overhead increases

Engineering Contradiction:

Improvenetwork efficiencyVSAvoidcontrol message overhead

Core Design Contradiction:

ProductivityVSQuantity of substance

Solution Approach 1:

The patent implements self-service by enabling downstream devices to autonomously make congestion isolation decisions based on local buffer status monitoring. Instead of requiring continuous upstream-downstream communication for every congestion event, the downstream device independently identifies congested flows and directs them to isolated queues, reducing control message overhead while maintaining isolation effectiveness.

Inventive Principle:

Principle #25Self-service

4Device complexity

If randomization is used to select flows for isolation, then complexity is reduced, but measurement precision decreases

Engineering Contradiction:

Improveflow selection complexityVSAvoidcongestion identification accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent implements feedback mechanisms where downstream devices continuously monitor buffer occupancy levels in real-time. This feedback allows the system to dynamically identify which flows are causing congestion based on actual buffer status, compensating for the lack of precise flow identification in random selection by using ongoing monitoring to detect and respond to congestion patterns.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS11349761B2Cost effective congestion isolation for lossless ethernet

Publication Date: 2022.05.31 HEWLETT PACKARD ENTERPRISE DEV LP

AI summary

A cost effective congestion isolation for lossless Ethernet utilizing random selection is provided. Upon detecting congestion at one of its egress ports, a downstream device randomly selects a data packet received at one or more of its ingress ports for isolation in a congested priority queue at the ingress port. The downstream device sends a congestion isolation message, including a list of isolation hashes to an upstream device to identify data packets selected for isolation. Upon receipt, the upstream device identifies data packets selected for isolation and begins to queue such packets in a congested priority queue of the upstream device. When priority-based flow control is triggered, the upstream device ceases transmission of packets in the congested priority queue only, while allowing continued transmission of non-isolated data packets.