Congestion-Aware Load Balancing in Data Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current load balancing methods in data networks, such as ECMP, are congestion-agnostic and inefficient, leading to poor performance and bandwidth utilization, especially in multi-tier networks with dynamic congestion levels and asymmetrical topologies, which hinders scalability and data integrity.

Innovation Solution

A congestion-aware load balancing method that employs a two-stage path selection process and local congestion monitoring to determine the optimal path based on real-time congestion information across network switches, using dedicated headers to transmit and process congestion data, and maintaining path selection tables for efficient routing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If ECMP (Equal Cost Multi-Path routing) is used for load balancing, then path selection is simplified and implementation is easier, but congestion awareness is lost leading to poor bandwidth utilization and performance

Engineering Contradiction:
Improvepath selection implementationVSAvoidbandwidth utilization
Core Design Contradiction:
Ease of operationVSProductivity

Solution Approach 1:

The patent implements feedback by having destination switches send congestion information back to source switches through dedicated header fields in returning packets. Source switches collect congestion metrics from multiple paths and use this feedback to make informed load balancing decisions, thereby achieving both ease of implementation and improved bandwidth utilization through congestion-aware path selection.

Inventive Principle:
Principle #23Feedback

2Measurement precision

If congestion monitoring is implemented across all network switches, then path selection accuracy is improved, but system complexity and resource consumption increase

Engineering Contradiction:
Improvecongestion detection accuracyVSAvoidnetwork switch complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by having each switch monitor congestion only on its directly connected uplinks rather than the entire network. Each switch maintains local congestion state information for its own ports, and this localized information is aggregated by source switches to make path selection decisions. This approach achieves accurate congestion detection without requiring complex global network state tracking at each switch.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent makes existing packet structures multi-functional by using dedicated header fields that serve dual purposes: carrying routing information and conveying congestion metrics. Return packets not only deliver data but also transport congestion information from destination to source switches. This universal use of packet fields reduces the need for separate monitoring mechanisms and minimizes additional system complexity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Manufacturing precision

If real-time congestion information is collected and processed, then path selection optimality is improved, but processing time and latency increase

Engineering Contradiction:
Improvepath selection optimalityVSAvoidpath selection latency
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent implements preliminary action by having destination switches prepare congestion information in advance and embed it in returning packets before source switches need to make path selection decisions. Congestion metrics are collected and made available proactively rather than on-demand, allowing source switches to quickly query and use pre-computed congestion data for immediate path selection without time-consuming measurements.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses returning data packets as intermediaries to transport congestion information from destination to source switches. Instead of requiring direct communication channels or separate control messages, the existing bidirectional data flow serves as the medium for congestion information exchange. This approach leverages already-necessary packet transmissions to carry additional routing information without adding separate communication overhead.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Measurement precision

If dedicated header fields are used to transmit congestion information, then information accuracy is improved, but packet size and overhead increase

Engineering Contradiction:
Improvecongestion information accuracyVSAvoidpacket data volume
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent makes existing packet structures multi-functional by using dedicated header fields that serve dual purposes: carrying routing information and conveying congestion metrics. Return packets not only deliver data but also transport congestion information from destination to source switches. This universal use of packet fields reduces the need for separate monitoring mechanisms and minimizes additional system complexity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent changes the state or form of data by encoding congestion metrics in a compact format within header fields. Rather than transmitting raw measurement data or detailed statistics, the system uses efficient parameter representations that convey essential congestion state information in minimal space. This allows accurate congestion information transmission while maintaining small packet overhead.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS10454830B2System and method for load balancing in a data network
Publication Date: 2019.10.22 CITY UNIVERSITY OF HONG KONG
  • US10454830B2 patent drawing
  • US10454830B2 patent drawing
  • US10454830B2 patent drawing

AI summary

A method for load balancing in a data network including a plurality of network switches includes receiving a data flow, initiating a path selection process to determine a path between a source network switch and a destination network switch, the path selection process selecting an optimal path based on congestion information of the first network switch and the congestion information of the second network switch, the optimal path corresponding to a path of minimal congestion, and transmitting the data flow along the selected path.