Edge Node Scaling via Deterministic Flow Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Logical networks in datacenters face bottlenecks due to the active-standby configuration of stateful gateways, where only one edge node is active at a time, leading to inefficiencies in handling 'north-south' traffic.
Innovation Solution
Implementing an active-active configuration for edge nodes using equal-cost multi-path (ECMP) algorithms and deterministic algorithms like consistent hashing to distribute data message flows across multiple active edge nodes, with a staged transition process to scale the number of edge nodes, allowing data messages to be processed by different nodes in both directions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If active-standby configuration is used for stateful gateways, then reliability is improved through failover capability, but productivity deteriorates due to single node processing bottleneck
Solution Approach 1:
The patent segments the single gateway function across multiple edge nodes, dividing the traffic handling responsibility. Each edge node processes a subset of data message flows, transforming the monolithic active-standby architecture into a distributed active-active system where N nodes handle N portions of traffic simultaneously, eliminating the single-node bottleneck while maintaining failover capability
Solution Approach 2:
The patent implements dynamic flow assignment using consistent hashing algorithms that automatically adapt to changes in the number of edge nodes. When nodes are added or removed, the system dynamically recalculates flow distribution without manual reconfiguration, enabling the gateway cluster to flexibly scale while maintaining balanced load distribution and continuous operation
2Productivity
If active-active configuration is implemented to increase productivity, then traffic handling capacity is improved, but device complexity increases due to state synchronization requirements
Solution Approach 1:
The patent extracts the stateful processing function from the data plane and relocates it to the control plane through centralized flow tables. Edge nodes forward packets based on distributed flow tables that contain state information, separating the high-speed forwarding function from the complex state management function. This extraction eliminates the need for complex inter-node state synchronization while maintaining active-active operation
Solution Approach 2:
The patent introduces flow tables as an intermediary data structure that mediates between multiple edge nodes and the stateful processing requirements. The flow tables, distributed across edge nodes using consistent hashing, store flow state information locally, allowing each node to independently process its assigned flows without requiring real-time synchronization with other nodes, thus reducing system complexity
3Productivity
If ECMP algorithms are used to distribute traffic across multiple edge nodes, then productivity is improved through load balancing, but measurement precision deteriorates due to difficulty in tracking flow state
Solution Approach 1:
The patent performs preliminary flow classification and state table assignment before actual data message processing. Using consistent hashing on flow identifiers (such as 5-tuples), the system pre-determines which edge node will handle each flow and pre-allocates the corresponding state table entry. This preliminary action ensures that subsequent packets in the same flow are directed to the correct node with the correct state information already in place, maintaining both load balancing and flow state tracking precision
Solution Approach 2:
The patent implements feedback mechanisms where edge nodes monitor their assigned flow states and report back to the control plane. The control plane uses this feedback to verify flow distribution correctness, detect imbalances, and dynamically adjust flow table assignments. This closed-loop feedback system ensures that ECMP-based load balancing maintains accurate flow state tracking even as traffic patterns change
Data Source
AI summary
Some embodiments provide a method for forwarding data messages between edge nodes that perform stateful processing on flows between a logical network and an external network. At a particular edge node, the method receives a data message belonging to a flow. The edge nodes use a deterministic algorithm to select one of the edge nodes to perform processing for each flow. The method identifies a first edge node to perform processing for the flow in a previous configuration and a second edge node to perform processing for the flow in a new configuration according to the algorithm. When the first and second edge nodes are different, the method uses a probabilistic filter and a stateful connection tracker to determine whether the flow existed prior to a particular time. When the flow did not exist prior to that time, the method selects the second edge node for the received data message.