Clos Network Path Congestion Detection and Switching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Clos data center network architecture experiences severe path congestion due to defects in conventional load balancing mechanisms, leading to packet loss and inefficiencies, especially in large-scale networks like those used by social media giants.
Innovation Solution
A transmission path determining method that assesses path congestion using a congestion information table, switching packets only when the current path is truly congested, and selecting alternative paths based on parameters like ECN, RTT, and flow rates to avoid congestion mismatch and long-tail effects, without requiring modifications to network devices like switches or routers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional load balancing mechanism is used in Clos architecture, then network scalability is improved, but path congestion occurs leading to packet loss
Solution Approach 1:
The patent implements a feedback mechanism where the sending host continuously monitors path congestion status by detecting ECN marks on returned packets and measuring RTT variations. Based on this feedback, the host dynamically adjusts path selection for subsequent packets, switching from congested paths to alternative less-congested paths. This closed-loop feedback system resolves the contradiction by enabling the network to maintain reliability through adaptive path selection while preserving the scalability of Clos architecture.
Solution Approach 2:
The patent transforms the static path selection of conventional load balancing into a dynamic system. The sending host continuously evaluates multiple available paths based on real-time congestion indicators (ECN marks, RTT measurements) and adaptively selects the optimal path for each packet or packet batch. This dynamic adaptation allows the system to maintain high packet transmission reliability while leveraging the scalable Clos network structure.
2Reliability
If path switching is performed frequently to avoid congestion, then packet loss is reduced, but congestion mismatch and long-tail effects occur
Solution Approach 1:
The patent applies partial action by switching paths only when necessary - specifically when congestion indicators (ECN marks or RTT thresholds) are exceeded. Rather than continuously switching paths or switching on minor fluctuations, the system performs path switching only when clear congestion evidence exists. This selective, partial switching approach reduces unnecessary path changes that cause congestion mismatch and long-tail delays, while still maintaining high packet transmission success rates by switching when truly needed.
3Measurement precision
If congestion information is monitored continuously, then path selection accuracy is improved, but system complexity increases
Solution Approach 1:
The patent implements self-service congestion monitoring where the sending host autonomously measures path congestion status using readily available network protocols. The host sends probe packets, measures RTT by timing packet round-trips, and detects ECN marks returned by network devices - all without requiring external monitoring infrastructure. This self-service approach achieves precise congestion detection while avoiding the complexity of dedicated monitoring systems, as the host leverages existing TCP/IP stack capabilities and standard network device features.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Embodiments of this application provide a transmission path determining method and apparatus, where the method includes: determining that a current path corresponding to a flow to which a to-be-transmitted packet belongs is congested; and determining a target path for the to-be-transmitted packet based on a path congestion information table, and adding information about the target path to the to-be-transmitted packet, so as to transmit the to-be-transmitted packet based on the target path, where a congestion degree of the target path is less than a congestion degree of the current path, each entry of the path congestion information table includes a transmission path and congestion information corresponding to the transmission path, and the congestion information is used to indicate a congestion degree of the transmission path. This effectively resolves a problem of transmission path congestion.