How Alibaba Scaled Its E‑Commerce: From LAMP to Distributed Multi‑Active Architecture

LAMP Architecture

Initially Alibaba's e‑commerce platform was built on a LAMP stack, with PHP front‑end and MySQL database, using read/write splitting after initial development.

Java Architecture

Growth forced migration to Java; Oracle replaced MySQL, but connection‑pool issues led to a phased replacement with Java services, introducing search components and moving from 1.0 to 2.0 architecture.

Distributed Architecture

Increasing traffic required distributed solutions: memory cache (TAIR), distributed file system (TFS), and a middleware layer to alleviate database pressure.

Multi‑Active (Active‑Active) Architecture

To address resource limits, scalability, and disaster recovery, Alibaba adopted a multi‑active, globally distributed architecture, partitioning business into logical units (C‑class, P‑class, vertical teams) and implementing global routing, eventual consistency, and cross‑unit isolation.

Key Middleware and Tools

Developed HSF RPC framework, Pandora container for jar isolation, ConfigServer for service discovery, TDDL for database sharding, “Yugong” for Oracle‑to‑MySQL migration, and “Jingwei” for change‑data capture and cache invalidation. Also built EagleEye for distributed tracing and a unified messaging system for asynchronous processing.

High Availability and Capacity Planning

Implemented full‑link stress testing, Sentinel for flow control, and elastic scaling platforms (Athena, Zeus) to ensure high availability during peak events like Double‑11.

New Starting Point

Alibaba now offers its architecture as Aliware on the cloud, providing enterprise‑grade services such as DRDS, EDAS, and MQ to help other businesses accelerate their growth.