Flow Size Classification Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional flow size classification methods, such as binary mice/elephant classification, are insufficient as they treat small and large elephant flows similarly, leading to excessive costs in latency, bandwidth, and resource usage, and do not provide meaningful size classifications.
Innovation Solution
An end-to-end data-driven approach using a configurable artificial intelligence engine processes packet data to determine multiple meaningful flow size classes based on bandwidth, allowing for more precise classification and efficient resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If binary mice/elephant classification is used, then classification simplicity is maintained, but classification precision deteriorates because small and large elephant flows are treated similarly
Solution Approach 1:
The patent segments flows into multiple size classes (e.g., mice, small elephant, large elephant) rather than using a single binary classification. This is achieved by implementing a flow classifier that outputs multiple discrete size categories, allowing differentiated handling of flows with different bandwidth characteristics while maintaining structured classification management.
Solution Approach 2:
The patent changes the classification parameter from a simple binary state to a multi-level categorical parameter representing different flow size classes. The flow classifier determines which of several predefined size classes a flow belongs to based on bandwidth measurements, enabling more nuanced flow identification and subsequent differentiated networking treatments.
2Manufacturing precision
If multiple flow size classes are implemented, then resource allocation precision is improved, but system complexity increases
Solution Approach 1:
The patent performs preliminary classification of flows into size classes before resource allocation decisions are made. The flow classifier pre-determines the size class of each flow based on initial bandwidth measurements, enabling downstream networking components to apply appropriate resource allocation policies without performing complex real-time analysis.
Solution Approach 2:
The patent introduces a flow classifier as an intermediary component between packet reception and resource allocation. This mediator component simplifies the overall system by centralizing the classification logic and providing standardized size class labels that other components can use for their respective functions without needing to implement their own classification mechanisms.
3Speed
If conventional binary classification is used, then processing speed is maintained, but resource utilization efficiency deteriorates due to excessive latency and bandwidth costs
Solution Approach 1:
The patent implements dynamic flow size class classification that adapts to different flow characteristics. The flow classifier continuously monitors bandwidth and adjusts size class assignments accordingly, enabling the system to optimize resource allocation in real-time based on actual flow behavior rather than using static binary thresholds.
Solution Approach 2:
The patent applies different resource allocation policies and handling mechanisms tailored to each flow size class. Small elephant flows receive different treatment compared to large elephant flows, with each class optimized for its specific characteristics. This local optimization improves overall resource utilization efficiency by matching handling strategies to actual flow needs rather than applying uniform binary classification rules.
Data Source
AI summary
A network device, system-on-a-chip, and method of performing packet handling are described. A packet is received, and data associated with the packet is processed, using a configurable artificial intelligence engine, to generate a size classification for a flow associated with the packet. An action is performed based, at least in part, on the size classification for the flow associated with the packet.