Offloading Service Mesh to DPU for Micro-service Governance

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In micro-services architectures, the current Service Mesh deployment mode leads to resource competition between sidecars and micro-service applications, affecting service performance and increasing deployment costs, as sidecars share resources like CPU, OS, and memory with service processes.

Innovation Solution

The proposed solution involves offloading the Service Mesh to a Data Processing Unit (DPU) to perform micro-service processing operations, achieving physical isolation and utilizing a combination of programmable hardware and software units to execute micro-service governance logic, thereby reducing resource competition and improving service performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If Service Mesh is deployed using sidecars in each micro-service application, then micro-service governance logic can be implemented, but resource competition occurs between sidecars and service processes affecting performance and increasing deployment cost

Engineering Contradiction:
Improvemicro-service governance capabilityVSAvoidservice performance
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system is segmented into two independent parts: micro-service applications running on computing nodes and Service Mesh running on separate DPU devices. This physical separation eliminates resource competition while maintaining governance functionality through network communication between the components.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The Service Mesh governance logic is extracted from the micro-service applications and deployed independently on DPU devices. This extraction removes the harmful resource competition while preserving the essential governance capabilities through dedicated hardware resources.

Inventive Principle:
Principle #2Taking out (Extraction)

2Reliability

If sidecars are deployed with each micro-service application to implement governance logic, then service governance functions are achieved, but deployment cost increases

Engineering Contradiction:
Improvegovernance function implementationVSAvoiddeployment cost
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

Multiple micro-service governance functions are merged into a single Service Mesh deployment on the DPU device. Instead of each application having its own sidecar, the governance capability is consolidated and shared across all micro-services through the unified Service Mesh implementation.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The Service Mesh deployed on the DPU provides universal governance functionality that serves all micro-service applications. This multi-functional approach eliminates the need for separate sidecars for each application, reducing deployment complexity while maintaining comprehensive governance coverage.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Use of energy by moving object

If sidecars share resources with micro-service applications, then resource utilization is shared, but resource competition affects service performance

Engineering Contradiction:
Improveresource sharingVSAvoidservice performance
Core Design Contradiction:
Use of energy by moving objectVSProductivity

Solution Approach 1:

Network communication acts as an intermediary between the micro-service applications and the Service Mesh on the DPU. This intermediary enables resource sharing through information exchange while preventing direct resource competition, as the two systems operate on separate hardware platforms.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS20250023942A1Information processing method, computer device and storage medium
Publication Date: 2025.01.16 BEIJING VOLCANO ENGINE TECH CO LTD
  • US20250023942A1 patent drawing
  • US20250023942A1 patent drawing
  • US20250023942A1 patent drawing

AI summary

An information processing method, a computer device, and a storage medium are provided. The method includes: acquiring communication information to be processed; extracting a key information character of the communication information to be processed by using a programmable hardware unit of a data processor; and according to the key information character, determining whether there is matching target processing operation information in a generated character mapping table, a character mapping table is generated by the programmable software unit of the data processor according to historical key information characters of historical communication information and their corresponding historical processing operation information, and the target processing operation information includes at least one micro-service processing operation; if yes, executing the micro-service processing operation included in the target processing operation information for the communication information to be processed.