How Ant Financial’s LDC Architecture Scales to 540k TPS on Double‑11

Background

Since 2008, Ant Financial’s Double‑11 traffic has grown from 20,000 payments per minute to 540,000 payments per second in 2019, a 1,360‑fold increase.

LDC and Unitization

LDC (Logical Data Center) treats distributed resources as a single logical entity, emphasizing unitization—splitting the system into independent units that each serve a specific user segment.

RZone, GZone, CZone

RZone handles user‑sharded data (isolated per user group), GZone stores globally shared data (single copy), and CZone stores data with a write‑read delay, allowing local reads while writes propagate asynchronously.

Traffic Routing and Disaster Recovery

Requests are routed by GLSB to the appropriate IDC based on IP, then Spanner determines the target RZone. If a unit fails, traffic is switched to a backup unit (same IDC) or to another IDC, with flow‑steering configurations updated before traffic cut‑over.

CAP Analysis

The system’s design addresses partition tolerance by using Paxos‑based consensus in OceanBase (OB). OB requires a quorum of (N/2)+1 nodes for writes, ensuring availability during partitions while achieving eventual consistency.

Database Layer

OB provides read/write capability per node, synchronizes writes to a quorum, and resolves split‑brain scenarios via consensus, thus satisfying AP with eventual consistency (C).

Conclusion

Key factors for Ant’s Double‑11 success are user‑sharded RZone design, OB’s Paxos‑driven consistency, CZone local reads for delayed‑write data, and robust disaster‑recovery mechanisms, enabling the platform to handle far higher TPS than current peaks.