Congestion-Aware Load Balancing via Probe Packets in Data Center Networks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Data center networks face challenges in load balancing due to uneven traffic distribution across paths, leading to congestion and performance degradation, as existing methods like ECMP are congestion-agnostic and lack flexibility, while control-plane and host-based approaches are slow or difficult to implement.

Innovation Solution

Implementing congestion-aware load balancing by switches in data center networks using probe packets to communicate congestion state information hop-by-hop, allowing switches to select the least congested path for data packet forwarding, thereby reducing congestion and improving network performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If ECMP (Equal Cost Multipath) is used for load balancing, then traffic is distributed across multiple paths, but congestion state is not considered leading to uneven traffic distribution and path over-utilization

Engineering Contradiction:

Improvenetwork throughputVSAvoidcongestion awareness

Core Design Contradiction:

ProductivityVSEase of operation

Solution Approach 1:

The patent implements feedback mechanisms where switches monitor congestion states on network paths and use this information to dynamically adjust load balancing decisions. Probe packets are sent to gather congestion state information from downstream switches, which is then fed back into the path selection process to avoid congested paths and distribute traffic more effectively.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The load balancing system performs self-service by having switches automatically monitor their own congestion states and make autonomous path selection decisions based on real-time conditions. Each switch independently gathers congestion information through probe packets and adjusts its forwarding behavior without requiring external control, enabling adaptive load balancing that responds to changing network conditions.

Inventive Principle:

Principle #25Self-service

2Adaptability or versatility

If control-plane load balancing approaches are used, then congestion awareness is improved, but implementation complexity and deployment difficulty increase

Engineering Contradiction:

Improvecongestion adaptabilityVSAvoidimplementation ease

Core Design Contradiction:

Adaptability or versatilityVSEase of manufacture

Solution Approach 1:

The patent introduces probe packets as intermediaries that carry congestion state information between switches. These probe packets serve as a mediator that enables congestion-aware load balancing without requiring complex control-plane protocols or modifications to end hosts. The probe packets simply traverse the network and convey necessary information for path selection.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The load balancing function is segmented into independent switch-level operations rather than requiring centralized control-plane coordination. Each switch independently performs congestion monitoring, probe packet processing, and path selection, dividing the complex task into manageable, distributed units that can operate autonomously and simplify deployment.

Inventive Principle:

Principle #1Segmentation

3Productivity

If host-based load balancing is implemented, then load distribution is improved, but processing overhead at end hosts increases and scalability is reduced

Engineering Contradiction:

Improvetraffic distribution efficiencyVSAvoidhost processing complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

Instead of having end hosts perform complex load balancing decisions, the patent inverts the approach by having network switches perform the congestion monitoring and path selection functions. The hosts simply send probe packets and receive responses, while the intelligent load balancing logic resides in the network infrastructure rather than at the endpoints, reducing host complexity and improving scalability.

Inventive Principle:

Principle #13The other way round (Inversion)

Solution Approach 2:

The patent replaces the mechanical approach of host-based processing with a network-based system where switches handle congestion monitoring and path selection. This substitution moves the computational burden from end hosts to network infrastructure, utilizing the switches' inherent capability to process and forward packets with minimal host involvement.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11777853B2Congestion-aware load balancing in data center networks

Publication Date: 2023.10.03 VMWARE INC

AI summary

Example methods are provided for a first switch to perform congestion-aware load balancing in a data center network. The method may comprise: receiving probe packets from multiple next-hop second switches that connect the first switch with a third switch via multiple paths. The method may also comprise: processing congestion state information in each probe packet to select a selected next-hop second switch from the multiple next-hop second switches, the selected next-hop second switch being associated with a least congested path from the first switch to the third switch. The method may further comprise: in response to receiving data packets from a fourth switch that are destined for a destination connected with the third switch, sending the data packets to the selected next-hop second switch such that the data packets travel to the third switch along the least congested path.