Inside Sina Weibo’s Third-Gen Backend Architecture for 100M+ Users

Introduction

In March 2014, Sina Weibo reported 143 million monthly active users and a record 808,298 posts in the first minute of the new year, demanding a highly available, high‑concurrency, low‑latency backend system.

First‑Generation Architecture

The initial platform used a LAMP stack: MyISAM for the database, PHP for the backend, and Memcache for caching.

Second‑Generation Architecture

As the service grew, the architecture shifted to a service‑oriented (SOA) model, replacing PHP with Java and modularizing business functions.

Third‑Generation Architecture

The current system employs an orthogonal decomposition model, forming a 3 × 4 matrix of horizontal and vertical layers.

Horizontal Layers

Interface Layer : Provides unified APIs for web pages and mobile clients, including core services such as Feed, User Relationship, and Messaging.

Service Layer : Divides services into atomic (e.g., short‑link, ID generator) and composite services (e.g., Feed, Messaging) built from atomic components.

Resource Layer : Manages data storage, including Redis, Memcache, MySQL, HBase, distributed file systems, and Sina S3.

Dependencies flow only downward: upper layers depend on lower ones, never the reverse.

Vertical Layers

The vertical dimension adds a technical foundation that supports the business architecture, monitoring platform, and service‑governance platform.

Key Technical Components

Interface Layer – Web V4 Framework

The framework standardizes API development using Spring AOP, builds on Jersey, and provides authentication, rate limiting, logging, degradation, and automatic serialization.

Service Layer Frameworks

MCQ Message Queue : A simple queue built on Memcache protocol with BerkeleyDB persistence, offering high performance and easy monitoring.

Motan RPC : Uses Netty as the transport, supports Hessian and Java serialization, and offers HA and load‑balancing strategies such as FailOver, FailFast, Round Robin, LRU, and Consistent Hash.

Resource Layer Frameworks

Object Store : Serializes objects to HBase with a unique ObjectID, supporting PB, JSON, and binary formats, and stores content in Sina S3.

SSDCache : Introduces an L2 cache (Redis/MC + SSD Cache + MySQL) to reduce cost, increase capacity, and alleviate database pressure.

Monitoring and Service Governance

To trace requests across distributed components, Weibo built the WatchMan tracing system, inspired by Google Dapper and Twitter Zipkin. It collects footprints, performance metrics, errors, and supports flow control, degradation, and automated scaling.

Conclusion

The platform’s evolving architecture—from LAMP to a sophisticated third‑generation stack—drives continuous technical upgrades, accelerates feature delivery, and underpins reliable operation at massive scale.