Congestion Notification Transfer via ARN Packets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network systems face challenges in efficiently managing and mitigating congestion through scalable and timely congestion notifications, particularly in large-scale computing environments like supercomputing and AI data centers, where current methods are not scalable and expensive.
Innovation Solution
A system and method for generating and propagating congestion notifications through a network, allowing switches to identify and resolve congestion without knowing the exact location of the relevant switch, using adaptive routing and timers to manage network traffic rerouting based on congestion notifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current congestion notification methods are used, then congestion can be detected, but the system becomes expensive and not scalable in large-scale computing environments
Solution Approach 1:
The patent extracts only the essential congestion notification function from the complex existing systems. Instead of implementing full-location-aware congestion management, the invention extracts and propagates minimal congestion indication information (ARN packets) through the network, allowing switches to respond to congestion without requiring knowledge of the exact congested switch location or implementing complex tracking mechanisms.
Solution Approach 2:
The patent introduces ARN (Adaptive Routing Notification) packets as intermediary messages that carry congestion information between switches. These ARN packets act as mediators that propagate congestion notifications through the network without requiring the source switch to know the exact location of congested switches, thereby simplifying the system architecture while maintaining congestion detection capability.
2Reliability
If switches propagate congestion notifications through the network, then congestion can be mitigated, but the notification process becomes complex and time-consuming
Solution Approach 1:
The patent segments the congestion notification process into discrete steps: detection at the congested switch, ARN packet generation, and propagation through intermediate switches. Each switch independently processes ARN packets and forwards them along the path to the source, dividing the complex notification task into manageable segments that simplify implementation at each node.
Solution Approach 2:
The patent implements feedback mechanisms where switches receive ARN packets containing congestion information and generate new ARN packets to forward the notification further along the network path. This feedback loop continues until the source switch receives the congestion notification, enabling automatic congestion mitigation without complex centralized control.
3Measurement precision
If the system tracks the exact location of congested switches, then precise congestion management is achieved, but the system requires more information and becomes less efficient
Solution Approach 1:
The patent applies partial action by having switches propagate ARN packets only as far as necessary to reach the source, without tracking the exact location of every congested switch. The system uses sufficient (but not excessive) information - ARN packets containing enough data for switches to identify congestion and reroute - avoiding the information overhead of complete location tracking while maintaining effective congestion management.
Data Source
AI summary
Systems, devices, and methods are provided. In one example, a system is described that includes circuits to receive a packet associated with a source, determine a device that received the packet is congested, in response to determining the switch that received the packet is congested, generate a congestion notification, transfer the congestion notification to a second device, wherein the second device is between the device and the source, set a timer, and in response to the congestion in the device not being resolved when the timer ends, generate and send a new congestion notification.