Exploring Service Mesh in Large-Scale Microservice Architecture: SOFA Mesh Technical Selection and Design

This article is a translated record of the first Service Mesher community meetup, featuring Ant Financial’s Service Mesh evangelist Ao Xiaojian. The original PPT and video of the talk are linked for reference.

The presentation, titled “Service Mesh Exploration Path under Large‑Scale Microservice Architecture,” emphasizes the challenges of performance, scalability, and high availability when deploying Service Mesh at Ant Financial’s massive scale.

Technical selection considered three main requirements: high performance, stability, and diverse deployment scenarios (internal main site, financial cloud, external customers) across heterogeneous environments (k8s, VMs, containers, public and private clouds). Two routes were evaluated – forking an existing open‑source product or building a new solution from scratch. Options such as Linkerd, Envoy, Istio, and Conduit were discussed, with reasons for rejecting each.

The final strategy adopts Envoy’s data‑plane concepts but rewrites the sidecar in Go, merges part of Mixer functionality into the sidecar, and enhances Pilot and Auth components. Check and quota logic are moved into the sidecar to avoid costly remote calls, while report remains in Mixer.

Architecture design details include a Go‑based sidecar that implements the XDS API, supports HTTP/1.1, HTTP/2, and extensions for SOFA, Dubbo, and HSF protocols. Mixer’s three functions—check, quota, and report—are explained, with a plan to integrate check and quota into the sidecar and keep report separate.

The talk also covers caching challenges, the performance impact of synchronous remote calls, and the need for efficient isolation between sidecar and backend services.

SOFA Mesh’s open‑source strategy aims to release the product early, offering a multi‑layer cooperation model: foundational libraries, modular components (sidecar, enhanced Pilot, Edge Sidecar), and a full‑stack mesh solution. The speaker stresses community collaboration, especially within the Chinese ecosystem, and outlines the goal of providing a production‑ready mesh for Ant Financial and its financial cloud.

Community resources are provided, including the official website, upcoming meetups, and WeChat groups for further discussion and contribution.