Flow Entry Failover Using Dynamic Routing After Link Failure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional ECMP traffic balancing mechanisms in AI data centers lead to uneven traffic distribution, resulting in some links becoming congested while others remain idle, reducing network throughput and AI cluster performance.
Innovation Solution
A network-wide load balancing solution using intelligent routing and dynamic routing protocols to handle link failures, where forwarding flow entries are prioritized and dynamically adjusted based on link status and latency, ensuring uninterrupted AI training tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If ECMP hashing mechanism is used for traffic balancing, then traffic distribution is achieved, but uneven traffic distribution occurs leading to link congestion and packet loss
Solution Approach 1:
The patent implements a feedback mechanism where the controller monitors link status and flow status in real-time, and dynamically adjusts flow entries based on this feedback. When link failures or congestion are detected, the controller receives status information, determines affected flows, and updates flow entries to redirect traffic, thereby resolving the uneven traffic distribution problem caused by static ECMP hashing.
Solution Approach 2:
The patent transitions from static ECMP hashing to dynamic flow steering. Flow entries are no longer fixed but are dynamically created, modified, and deleted by the controller based on real-time network conditions. This allows the system to adapt traffic distribution dynamically, preventing link congestion and improving both throughput and reliability.
2Speed
If flow entries are used for forwarding, then forwarding efficiency is improved, but link failures cause service interruption
Solution Approach 1:
The controller pre-calculates alternative forwarding paths and prepares flow entries for potential failover scenarios. When a link failure is detected, the controller can immediately activate pre-prepared flow entries to redirect traffic, minimizing service interruption. This preliminary preparation of alternative paths ensures both high forwarding speed and service continuity.
Solution Approach 2:
The system establishes backup flow entries and alternative routing paths in advance as a cushion against potential link failures. These pre-positioned protective measures ensure that when failures occur, traffic can be rapidly redirected without significant service interruption, maintaining both forwarding efficiency and reliability.
3Adaptability or versatility
If dynamic routing protocol is enabled, then routing adaptability is improved, but routing convergence time increases during link failures
Solution Approach 1:
The controller acts as an intermediary between the dynamic routing protocol and the forwarding devices. Instead of relying solely on slow routing protocol convergence, the controller receives link status notifications, quickly determines affected flows, and directly updates flow entries in forwarding devices. This intermediary approach bypasses the slow routing convergence process while maintaining routing adaptability.
Solution Approach 2:
The patent replaces the traditional mechanical routing protocol-based path selection with a controller-driven flow entry management system. The controller uses direct monitoring and control plane instructions to rapidly adjust traffic paths, substituting the slow, distributed routing protocol convergence mechanism with a centralized, faster control approach that maintains adaptability while reducing convergence time.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The present disclosure relates to a technical field of intelligent computing centers, and particularly to methods and apparatuses for handling a link failure, a storage medium, and a program product. The method is performed by a target network apparatus in a forwarding path in a networking to forward a target flow based on flow entries; where each of network apparatuses in the networking advertises or forwards a host route of a computing resource accessed by the networking to other network apparatuses in the networking based on a dynamic routing protocol; where the method includes: receiving a first flow entry sent by a controller, where the first flow entry is for forwarding the target flow, and the first flow entry includes a source network address, a destination network address, an egress interface, and information of a next hop apparatus of the target flow; monitoring a status of a link associated with the first flow entry; deleting the first flow entry in response to detecting that the link associated with the first flow entry faults; and forwarding the target flow based on the destination network address of the target flow and a dynamic routing entry.