Evolution of Taobao's Backend Architecture from Single‑Server to Cloud‑Native Microservices

This article uses Taobao as a case study to illustrate the step‑by‑step evolution of a high‑concurrency backend system, describing the technical challenges and solutions at each stage.

Basic concepts such as distributed systems, high availability, clustering, load balancing, and forward/reverse proxy are introduced to ensure readers understand the terminology.

Architecture evolution includes:

1) Single‑machine architecture where Tomcat and the database share one server, which quickly becomes a bottleneck as traffic grows.

2) Separation of Tomcat and the database onto different machines, improving resource isolation but exposing database read/write limits.

3) Introduction of local and distributed caches (e.g., Memcached, Redis) to offload read traffic from the database.

4) Deployment of reverse‑proxy load balancers (Nginx/HAProxy) to distribute requests across many Tomcat instances.

5) Database read/write splitting using middleware like Mycat to alleviate write‑side pressure.

6) Business‑level database sharding to reduce contention between different services.

7) Horizontal table splitting (small tables) to enable massive parallel processing (MPP) with databases such as TiDB, Greenplum, etc.

8) Use of LVS or hardware load balancers (F5) for layer‑4 traffic distribution and high‑availability via keepalived.

9) DNS round‑robin across data‑centers for global load balancing.

10) Integration of NoSQL stores and search engines (HBase, Redis, Elasticsearch, Kylin, Druid) for large‑scale data, analytics, and full‑text search.

11) Extraction of common functions into microservices (e.g., user, order, payment) using frameworks like Dubbo or Spring Cloud.

12) Adoption of an Enterprise Service Bus (ESB) to unify service interfaces and reduce coupling (SOA).

13) Containerization with Docker and orchestration via Kubernetes to isolate runtime environments and enable rapid scaling.

14) Migration to a cloud platform (IaaS/PaaS/SaaS) for on‑demand resource provisioning, reducing operational cost and supporting massive traffic spikes.

The article concludes with a set of architectural design principles, including N+1 redundancy, rollback capability, feature toggles, built‑in monitoring, multi‑active data‑center design, use of mature technologies, resource isolation, horizontal scalability, purchasing non‑core components, commercial hardware, rapid iteration, and stateless service design.

Overall, the piece provides a comprehensive roadmap for scaling backend systems from modest traffic to millions of concurrent users, highlighting both the technologies involved and the practical considerations for each evolutionary step.