Distributed Layer-2 Routing for East-West Traffic Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In cloud computing data centers, the increased east-to-west traffic leads to bottlenecks and limited bandwidth due to the need for all packets to pass through physical or virtual gateways, resulting in a sharp drop in average network bandwidth as concurrent communications preempt bandwidth from each other.
Innovation Solution
Implementing a distributed routing method where a controller determines if two servers are in different subnets under the same router and creates distributed routing rules to modify packet headers at the layer-2 data link layer, allowing packets to bypass layer-3 routers and be forwarded directly between switches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If all packets pass through physical gateway or virtual gateway (layer-3 routing), then routing control is centralized and manageable, but transmission bandwidth for east-to-west traffic is limited and becomes a bottleneck
Solution Approach 1:
The patent segments the routing function from the gateway and distributes it to switches. Instead of all packets passing through the gateway for routing decisions, switches locally execute routing rules to forward packets directly between servers. This segmentation eliminates the gateway bottleneck while maintaining controlled routing through pre-distributed rules.
Solution Approach 2:
The patent changes the routing dimension by implementing layer-2 distributed routing alongside traditional layer-3 gateway routing. This creates a dual-path architecture where east-to-west traffic can use the faster layer-2 switch-based routing for direct server communication, while layer-3 gateway routing handles other scenarios, thus increasing overall transmission bandwidth.
2Productivity
If all packets are processed at layer-3 routing protocol layer through gateway, then routing decisions are centralized, but network efficiency drops due to gateway bandwidth limitations and concurrent communication preemption
Solution Approach 1:
The patent applies preliminary action by pre-distributing routing rules to switches before traffic flows. The controller calculates optimal routing rules based on network topology and server locations, then distributes these rules to relevant switches in advance. This eliminates real-time routing decision delays at the gateway, as switches can directly forward packets based on pre-computed rules.
Solution Approach 2:
The patent enables packets to skip the gateway processing step by implementing direct layer-2 switching between source and destination servers. When switches have the appropriate distributed routing rules, packets are forwarded directly at layer-2 without being sent to the layer-3 gateway, thus skipping the bandwidth-constrained gateway path and reducing transmission delay.
3Productivity
If physical gateway or virtual gateway is used for cross-subnet communication, then routing control is simplified, but east-to-west traffic bandwidth is constrained by gateway capacity
Solution Approach 1:
The patent implements self-service by enabling switches to autonomously forward packets using locally stored distributed routing rules. Instead of requiring centralized gateway processing for every packet, switches independently make forwarding decisions based on pre-distributed rules that contain information about server locations and optimal paths, thus increasing east-to-west bandwidth without requiring complex manual routing configuration.
Solution Approach 2:
The patent introduces the controller as an intermediary that automatically generates and distributes routing rules to switches. This intermediary role simplifies the overall system by centralizing the complex routing rule calculation and distribution task, while allowing switches to operate independently with simple local rules, thus maintaining ease of operation through automated rule management.
Data Source
AI summary
A method for packet routing in a network is disclosed. The network includes a network controller, a first switch connected to the network controller, a first server directly connected to the first switch, and a second server. The network controller obtains network connection information of the two servers. The network connection information indicates that the first server belongs to a first subnet and the second server belongs to a second subnet, the first subnet and the second subnet are two different subnets, and the two different subnets are configured on a same router. The network controller sends configuration information to the first switch. The configuration information configures the first switch to modify a packet header of a packet sent from the first server to the second server, and forward the modified packet to the second server.