Rethinking Future Database Architecture: Insights from Alibaba’s Lead Engineer

Speaker: Zhang Rui, head of Alibaba Group Database Technology team, Alibaba researcher, Oracle ACE, and chief architect of the Double‑11 database technology.

Overview of the Talk

At the 2017 China Database Technology Conference, Zhang presented his thoughts on the future of database system architecture, sharing Alibaba’s experiences, lessons learned, and forward‑looking ideas to help advance China’s database technology.

Problems Alibaba’s Databases Face

Alibaba’s workloads are transaction‑heavy, requiring strong consistency and high availability; storage costs are high; data exhibits hot‑cold patterns; and events like Double‑11 create extreme load spikes that demand extreme elasticity.

Kernel Innovations

AliSQL X‑Cluster : A three‑node cluster built on AliSQL that incorporates the Paxos consensus protocol to achieve strong consistency, multi‑region deployment, and tolerance of high network latency.

Paxos is used for leader election and to guarantee strong consistency without shared storage, enabling cross‑datacenter replication.

X‑KV : An enhanced Memcached plugin for MySQL that provides ultra‑fast key‑value access directly from the InnoDB buffer, reducing the need for separate caching layers.

Dual‑Engine Storage : Combining InnoDB (B‑Tree) for balanced OLTP workloads with RocksDB (LSM‑Tree) for write‑friendly, high‑compression scenarios, allowing hot‑cold data separation and cost‑effective storage.

Why Database Elastic Scheduling Matters

During Double‑11, the need for elastic scaling of databases becomes critical; achieving this requires containerization and separation of compute and storage so that databases can be moved and scaled like applications.

The Future Role of DBAs

Alibaba is moving from manual, rule‑based DBA tasks toward automation and intelligent optimization, collecting every SQL execution, integrating application‑side metrics, and building a monitoring‑driven optimization engine that leverages large‑scale data and machine learning.

Other Reflections on Database Evolution

There is no single database that solves all problems; a layered platform approach—separating a support platform that abstracts storage complexity from a service platform that offers unified database services—helps reduce DBA effort and improves scalability.

Automation brings paradoxes: while it reduces manual work, it can also erode human expertise, so balancing system automation with human capability is essential.