Active-Active Stateful Network Service Cluster Traffic Redirection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Stateful network services face bottlenecks in scaling due to high CPU intensity, particularly in virtualized environments, where traditional active-active clustering methods lead to significant traffic redirection, reducing throughput and increasing CPU load.

Innovation Solution

A method where a first service node identifies a second service node to process data messages and forwards state information for subsequent messages, allowing efficient routing and minimizing traffic redirection by using an overlay network and consistent hashing to manage service node memberships and state synchronization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If consistent hashing is used to select a specific node for each session in active-active clustering, then session state information can be maintained, but traffic redirection significantly lowers performance and scalability

Engineering Contradiction:
Improvesession state information maintenanceVSAvoidthroughput and scalability
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent applies preliminary action by pre-distributing session state information to multiple candidate nodes using consistent hashing before traffic arrives. When a node receives traffic for a session, it checks if it has the state information locally; if not, it retrieves it from a predetermined backup node identified through the consistent hashing function. This eliminates the need for real-time traffic redirection to the owning node, thereby maintaining high throughput while ensuring session state availability.

Inventive Principle:
Principle #10Preliminary action

2Ease of operation

If ECMP-based routing is used to distribute traffic among cluster nodes, then load balancing is achieved, but most traffic must be redirected to the correct node for processing

Engineering Contradiction:
Improveload balancingVSAvoidtraffic redirection overhead
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary mechanism where each node maintains a local cache or mapping table that translates ECMP-distributed traffic into the correct destination node for session processing. When a node receives ECMP traffic, it consults this intermediary structure to determine if it should process the session locally or forward it to the appropriate node. This intermediary layer decouples the simple ECMP distribution from the complex session state location, reducing unnecessary traffic redirection while preserving load balancing benefits.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Productivity

If vertical scaling is used to increase CPU capacity for stateful network services, then performance improves, but the cost is very expensive and CPU capacity is limited by physical server constraints

Engineering Contradiction:
ImproveCPU capacity and performanceVSAvoidcost and scalability
Core Design Contradiction:
ProductivityVSEase of manufacture

Solution Approach 1:

The patent applies segmentation by dividing the stateful network service functionality across multiple independent virtual instances running on shared physical infrastructure. Each virtual instance maintains a portion of the session state information and can process traffic independently. This segmentation allows horizontal scaling where additional virtual instances can be created on existing hardware or distributed across multiple servers, avoiding the need for expensive vertical scaling while maintaining stateful service capabilities through the consistent hashing-based state distribution mechanism.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS11570092B2Methods for active-active stateful network service cluster
Publication Date: 2023.01.31 VMWARE INC
  • US11570092B2 patent drawing
  • US11570092B2 patent drawing
  • US11570092B2 patent drawing

AI summary

For a managed network, some embodiments provide a method for a set of service nodes in an active-active service node cluster in conjunction with a host computer hosting a destination data compute node (DCN) to improve the efficiency of directing a data message to a service node storing state information for the flow to which the data message belongs. a first service node receives a data message in a particular data message flow for which it does not maintain state information. The first service node then identifies a second service node to process the data message and forwards the data message to the second service node. The second service node sends state information for the particular data message flow to the first service node, for the first service node to use to process subsequent data messages in the particular data message flow.