Heavy Hitter Flow Detection in Programmable Packet Pipelines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In datacenter and ISP networks, heavy hitter flows account for a majority of traffic, leading to network congestion and increased latency, necessitating efficient detection and mitigation strategies to manage these flows effectively.
Innovation Solution
A packet processing system that identifies candidate heavy hitter flows through monitoring traffic volume and uses a voting mechanism to determine and manage these flows, reducing memory resource utilization and implementing actions such as queuing and flow rate control to mitigate congestion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional flow detection methods are used to monitor all packet flows, then detection accuracy is improved, but memory resource utilization increases significantly
Solution Approach 1:
The patent segments the flow detection problem by dividing flows into two categories: heavy hitter flows (monitored in detail) and non-heavy hitter flows (monitored coarsely). This segmentation allows the system to allocate memory resources efficiently by maintaining detailed state information only for flows that consume significant bandwidth, while using compact representations for the majority of flows that consume less bandwidth.
Solution Approach 2:
The patent applies local quality by using different monitoring granularities for different flows. Heavy hitter flows receive detailed monitoring with full packet header information and precise byte counting, while non-heavy hitter flows receive coarse monitoring with aggregated statistics. This local differentiation optimizes memory usage by applying high-precision tracking only where necessary.
2Reliability
If heavy hitter flows are not detected and managed, then network operates normally, but network congestion and latency increase
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously monitors flow characteristics and dynamically adjusts monitoring and management actions. When a flow exceeds threshold bandwidth consumption, the system triggers heavy hitter detection and applies mitigation actions such as rate limiting or queue prioritization. This feedback loop ensures network performance is maintained while avoiding unnecessary complexity for normal operating conditions.
Solution Approach 2:
The system employs self-service mechanisms where flows that exceed bandwidth thresholds automatically trigger detection and management protocols. The heavy hitter detection algorithm autonomously identifies problematic flows and applies appropriate mitigation strategies without requiring manual intervention, thereby maintaining network reliability while managing complexity through automation.
3Measurement precision
If detailed monitoring of all flows is implemented, then congestion detection accuracy is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by implementing detailed monitoring only for a subset of flows identified as potential heavy hitters, rather than monitoring all flows with equal detail. The system uses initial coarse monitoring to identify candidates, then applies excessive (detailed) monitoring only to those candidates, thereby achieving accurate congestion detection while minimizing overall processing time.
Solution Approach 2:
The system performs preliminary action by conducting initial coarse-grained monitoring of all flows to identify potential heavy hitters before applying detailed monitoring. This preliminary filtering step reduces the number of flows requiring intensive processing, thereby achieving accurate congestion detection for critical flows while minimizing total processing time through staged analysis.
Data Source
AI summary
Examples described herein relate to a programmable packet processing pipeline configured to: access a data corresponding to multiple bins, respective bins associated with multiple packet flows and for respective bins: identify a single flow associated with a bin of the multiple bins as a candidate heavy hitter flow and determine a different packet flow as the candidate heavy hitter flow for the bin of the multiple bins.