How Alibaba Cut Double‑11 Transaction Costs by 75% with Cloud‑Native Architecture

To mark the upcoming tenth Double 11, Alibaba Technology launched the "Ten‑Year Code Chronicle" series, inviting core engineers who supported past Double 11 events to review the platform's evolution.

Senior Platform Technology expert Uncle Tong describes three phases of infrastructure development: a technology‑catch‑up period (pre‑2012) focused on stability, a maturity period (2013‑2014) that introduced full‑link pressure testing and multi‑active deployments, and an explosion period (2015 onward) featuring containerization, unified scheduling, hybrid‑cloud elastic architecture, hybrid deployment, and storage‑compute separation.

Key milestones include:

2013: Full‑link pressure testing launched, enabling deterministic stability verification across the entire business chain.

2014: Unit‑level multi‑active architecture applied to Double 11, providing remote, cross‑region deployment and scaling capabilities.

These advances, combined with elastic resource strategies, reduced the incremental cost of ten‑thousand‑transaction spikes to one‑quarter of the 2012 level.

Cost reduction formula: Promotion resource cost = resource holding time × resource holding scale . By shifting from year‑round resource holding to elastic, on‑demand capacity, Alibaba cut the added cost per ten‑thousand transactions by 50% each year from 2015 to 2018.

Hybrid‑cloud elastic architecture, built on the internal Sigma scheduler and Alibaba Cloud resources, allows rapid provisioning of temporary transaction units via one‑click site creation, dramatically lowering peak‑capacity expenses.

Unified scheduling treats the entire group as a single cloud, enabling cross‑data‑center resource orchestration, mixed deployment of online services and batch compute, and fine‑grained isolation of CPU, memory, I/O, and network resources.

Storage‑compute separation (Pangu storage) abstracts compute and storage nodes, smoothing heterogeneity across hardware generations and supporting millisecond‑level adaptive scheduling.

Overall, containerization, orchestration, and hybrid‑cloud deployment have turned the data center into a "computer", achieving over 45% CPU utilization in daily operations and a 30% reduction in IT costs.

The continued deepening of unified scheduling and cloud‑native technologies aims to bring future promotion‑related resource costs close to zero.