Designing Scalable Large-Scale Internet Applications: Stateless Sessions, Caching, Service Splitting, Database Sharding, Asynchronous Communication, and Configuration Management

Stateless design is essential for horizontal scalability; the article describes how Taobao's session framework stores state in client cookies using a multi‑value cookie approach, eliminating server‑side session storage and enabling easy node addition.

Effective caching, illustrated with Taobao's Tair, reduces database load by using both read and write caches; examples include caching shop pages and product details to lower DB pressure.

Service decomposition (HSF) separates a monolithic system into functional sub‑systems, improving maintainability, horizontal scaling, and fault isolation, while also introducing challenges such as inter‑service communication and dependency management.

Database splitting (TDDL) progresses from a single DB to master/slave replication, vertical partitioning (separate databases per domain), and finally horizontal sharding, addressing read/write pressure and large table size issues.

Asynchronous communication (Notify) decouples services via message middleware, enhancing system scalability, availability, and response time compared to synchronous calls.

Unstructured data storage combines a distributed file system (TFS) for small binary objects and NoSQL solutions (Cassandra, HBase, Bigtable) for key‑value data, employing BASE consistency to favor availability.

Monitoring and alerting systems track both coarse‑grained metrics (CPU, memory, traffic) and fine‑grained usage (page PV, bandwidth), enabling automatic alerts for abnormal conditions.

Unified configuration management ensures consistent settings across many nodes, simplifying node addition/removal and reducing configuration errors.