What Drives the Architecture of Billion‑User Platforms? Lessons from Weibo

Understanding the Essence of Architecture

Before discussing the essence of architecture, the author reflects on the strategic importance and tactical insignificance of handling million‑scale traffic, using Uber’s order volume as an example to illustrate the magnitude of a ten‑million‑level system.

What Is Architecture?

Architecture is described as a framework that holds business logic and algorithms, akin to a clothes rack, and more abstractly as an abstraction of repetitive business and a foresight of future expansion.

Key Capabilities for Architects

Abstraction : Removing redundancy to improve reusability across functions, classes, services, and templates.

Classification : Decoupling objects, defining attributes and methods, and modularizing services in distributed systems.

Algorithm (Performance) : Enhancing system performance by optimizing CPU, memory, I/O, and network.

Illustrative Examples

Examples include MySQL sharding and templating, CDN acceleration, service‑oriented architecture, and message queues as classification mechanisms.

Weibo’s Overall Architecture

Weibo follows a three‑tier architecture: client (Web, Android, iOS), an interface layer that provides security isolation, traffic control, and platform differentiation, and a backend consisting of platform services, search, and big‑data processing.

As traffic grows from millions to billions of users, the architecture evolves from first‑generation (supporting millions) to second‑generation (tens of millions) and third‑generation (hundreds of millions to billions), requiring service decomposition, stateless interfaces, and extensive monitoring.

Design Principles

Use RPC components, messaging middleware for asynchronous decoupling and traffic shaping, and configuration management for graceful degradation.

Maintain statelessness at the interface layer by moving state to caching or storage.

Prioritize data‑layer design to avoid costly schema migrations.

Map physical team organization to logical technical architecture for efficient collaboration.

Understand the full request path, including DNS, load balancers, and VIPs, before reaching the interface layer.

Identify bottlenecks in CPU, memory, storage, or network, often at a single node.

Multi‑Level Dual‑Data‑Center Caching

Weibo employs a two‑level cache (L1 and L2) across two data centers for the feed service, with L1 handling high QPS and L2 providing capacity, reducing database load and supporting hot‑spot traffic.

Feed Storage Architecture

Posts are stored in MySQL with primary and secondary indexes; sharding is performed by user ID and time to separate hot and cold data, enabling efficient retrieval for timeline generation.

Distributed Service Tracing

A tracing system propagates a unique request ID across RPC calls, enabling end‑to‑end monitoring with minimal intrusion, supporting multiple languages and standardized logging.

Operational Practices for Massive Traffic

During peak events, Weibo employs degradation plans, full‑stack load testing at five‑times normal traffic, and a shared Docker cluster for rapid scaling.

Continuous Learning Path

To improve architecture skills, the author recommends mastering Java, the JVM, operating systems, design patterns, TCP/IP, distributed systems, data structures, and algorithms.