The Evolution of Taobao’s Technology Architecture: From Monolith to Distributed Systems

Today we focus on the most decisive architectural evolution in Taobao’s history – the transition from the third to the fourth stage – which fundamentally reshaped the platform and provides a blueprint for large‑scale internet companies.

The third stage faced multiple pressures: massive codebase (Denali) maintained by hundreds of engineers, diverse business requirements (users, shops, payments, etc.), Oracle’s connection‑pool and CPU limits, and the need for horizontal scaling of servers and data centers.

To address these issues, Taobao vertically split the Denali codebase by business domain, creating dozens of independent services such as UM (User Manager) and SM (Shop Manager). Interfaces were also split per business, and a custom high‑performance communication framework, HSF, was introduced to support various remote‑call patterns.

Simultaneously, a middleware matrix emerged, including distributed cache Tair, distributed file system TFS, asynchronous messaging Notify, the database routing layer TDDL, and a session framework, all designed to reduce pressure on the Oracle database.

On the database side, Taobao performed both vertical and horizontal sharding: core services kept Oracle, while non‑core services migrated to MySQL; horizontal partitioning and read/write separation were managed by TDDL, allowing transparent routing without application changes.

Additional measures included building the proprietary ISearch engine to offload product search from Oracle, establishing an independent CDN infrastructure, creating an operations platform (the “Hubble Telescope”) for monitoring and alerting, and implementing disaster‑recovery data centers to ensure continuous availability.

The overall evolution required substantial investment but resulted in a robust, scalable, and distributed architecture that underpins Taobao’s current performance and reliability.