Active Load Balancing for Elephant Flows in Network Switches

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Static load balancing techniques fail to respond adequately to dynamically changing network traffic patterns in data center networks due to server virtualization and high-bandwidth application loads, leading to inadequate responsiveness and potential overloading of network links.

Innovation Solution

The Active Load Balancing (ALB) technique considers both local and remote link utilization information to make load balancing decisions, focusing on 'elephant flows' that have a significant impact on the network, thereby optimizing network-wide load balancing without incurring excessive resource and bandwidth overhead.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If dynamic load balancing considers remote link utilization information for all flows, then network-wide load balancing performance is improved, but resource and bandwidth overhead increases significantly

Engineering Contradiction:

Improvenetwork-wide load balancing performanceVSAvoidresource and bandwidth overhead

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments network flows into two categories: elephant flows (large flows that consume significant bandwidth) and mouse flows (small flows). The system applies different load balancing strategies to each segment - considering remote link utilization information for elephant flows while using simpler local information for mouse flows. This segmentation resolves the contradiction by focusing computational resources on the most impactful flows.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Instead of collecting remote link utilization information for all flows (excessive action), the patent applies this action selectively only to elephant flows (partial action). This partial application of the complex load balancing mechanism to the most critical flows achieves sufficient network-wide performance improvement while avoiding the excessive overhead that would result from applying it universally.

Inventive Principle:

Principle #16Partial or excessive action

2Device complexity

If static load balancing techniques are used, then device complexity is low, but responsiveness to dynamically changing network traffic patterns is inadequate

Engineering Contradiction:

Improveload balancing mechanism complexityVSAvoidresponsiveness to changing traffic patterns

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent implements a dynamic load balancing mechanism that adapts to changing network conditions by continuously monitoring link utilization information and recomputing paths for elephant flows when needed. This dynamic approach maintains low complexity for mouse flows while introducing adaptability specifically where needed for elephant flows, resolving the contradiction between simplicity and responsiveness.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent applies different quality levels of load balancing to different flow types: simple local information-based balancing for mouse flows and comprehensive remote-aware balancing for elephant flows. This local quality differentiation allows the system to maintain low overall complexity while achieving high responsiveness for the most critical traffic patterns.

Inventive Principle:

Principle #3Local quality

3Reliability

If remote link utilization information is collected for all flows, then load balancing decisions are optimized, but bandwidth overhead and resource consumption increase

Engineering Contradiction:

Improveload balancing decision qualityVSAvoidbandwidth and resource consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent segments flows into elephant and mouse categories, applying remote information collection only to elephant flows. This segmentation ensures that high-quality load balancing decisions are made for the most bandwidth-intensive flows while avoiding the unnecessary bandwidth and resource consumption that would result from collecting remote information for all flows.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies remote link utilization information collection partially - only for elephant flows rather than all flows. This partial action achieves optimized load balancing decisions for the critical subset of flows that consume the majority of bandwidth, while avoiding the excessive resource consumption that would result from universal application.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS10892992B2Load balancing

Publication Date: 2021.01.12 HEWLETT PACKARD ENTERPRISE DEV LP

AI summary

Example implementations relate to load balancing network traffic flows in a network switching device. According to an example, in a network switching device, a method to load balance network traffic flows across links of the device includes identifying a flow of the network traffic flows as an elephant flow. In response to identifying the flow as an elephant flow, the method collects remote utilization information for the flow. The method recomputes a path for the flow based on local utilization information and the collected remote utilization information and sends the flow along the recomputed path.