Alibaba's Multi‑Active Data Center Architecture: Design, Challenges, and Lessons Learned

In the era of big data, Alibaba launched a cross‑region active‑active data‑center project before the 2014 Double‑11 shopping festival, led by Lin Hao (nickname Bi Xuan), a researcher in Alibaba's Technical Assurance Department responsible for performance and capacity architecture.

The project, internally called "unitization" and progressing through three stages (unitization, dual‑active, multi‑active), aims to run multiple transaction services in different data‑center locations, each handling traffic independently.

Initially, Alibaba experimented with cross‑region data‑centers around 2009‑2010, following a same‑city multi‑data‑center model that provided failover capability. However, traditional cold‑backup approaches proved too costly and unreliable for Alibaba's massive e‑commerce platform.

By 2013, rapid growth of e‑commerce, logistics, cloud, and big‑data workloads made a single‑city data‑center insufficient, prompting the decision to adopt multi‑active deployment to improve scalability and reduce risk.

During Double‑11, two data‑centers (Hangzhou and another city) each handled 50% of user traffic, with all user actions—including browsing, searching, ordering, and database writes—performed locally without cross‑region calls.

The advantages of multi‑active deployment include real‑time traffic handling, lower risk compared to cold backup, and the ability to maintain service continuity even if one site fails.

Key challenges identified were:

Latency: cross‑city communication adds delay, which can be amplified by the hundreds of service calls per page in a large system like Taobao.

Routing consistency: ensuring that a user’s requests are consistently routed to the same data‑center throughout the entire request chain (frontend, backend services, database) to avoid data inconsistency.

Data consistency and synchronization: achieving sub‑second data sync across sites for critical buyer‑related data while preventing conflicting writes, especially given multiple data dimensions (buyer, seller, product).

Cost control: balancing redundancy with acceptable overhead, especially when scaling from two to three data‑centers.

To address latency, Alibaba built a dedicated backbone network between cities and developed custom data‑synchronization mechanisms now offered as a service on Alibaba Cloud.

The interview also mentions upcoming talks at QCon Beijing 2015 where Alibaba engineers will share details on high‑availability standards, fault‑recovery time objectives, and the multi‑active architecture that enables sub‑minute failover.