OceanBase Unitization: Building the Next Generation of Online Map Applications

IEEE International Conference on Data Engineering (ICDE) is one of the top conferences in the database and data engineering field. The paper titled "OceanBase Unitization: Building the Next Generation of Online Map Applications" reports the design and deployment of OceanBase (OB) in the Gaode (Amap) map platform.

Introduction

With the rapid growth of data, distributed database systems have become mainstream for internet services. Traditional single‑host architectures face challenges in scaling and supporting both OLTP and OLAP workloads. OceanBase’s column‑store mechanism enables a more flexible engine for diverse scenarios.

Unitization Architecture

OceanBase adopts a unit‑based design that migrates from a single‑host to a multi‑region, multi‑host deployment. A "unit" is an independent entity that hosts all services and data required for a set of business operations. Units are distributed across all IDC sites, providing seamless failover and high availability, while routing requests to geographically close units to reduce latency.

The architecture comprises three zones: Global Zone (GZone) for non‑shardable data, Region Zone (RZone) for shardable data, and City Zone (CZone) as a read‑only replica of GZone to serve local RZones. Each RZone shard has five replicas using Paxos for strong consistency, and the system supports three‑city‑five‑center deployment for disaster recovery.

Key Design Features

Pre‑write log (WAL) using PALF (Paxos‑based Append‑only Log File System) for asynchronous, high‑throughput writes.

Multi‑Paxos replication protocol for efficient consensus and leader election.

Dynamic read/write unitization: read‑heavy workloads can be unitized, while write‑heavy workloads may use centralized writes to reduce synchronization overhead.

Case Studies

Case 1 – Strong Consistency Finance: Financial settlement services require strict consistency and cross‑region disaster recovery. OceanBase’s unitized deployment ensures zero data loss and sub‑8‑second recovery time.

Case 2 – Cloud Sync (OLTP): Cloud synchronization across devices demands high write throughput and low latency. OceanBase’s unitized design delivers high availability and efficient multi‑point writes.

Case 3 – Online Review Service (OLAP): Review services involve read‑heavy workloads with millisecond‑level response time requirements. OceanBase’s architecture provides fast reads and scalable storage for high‑precision map data.

Experimental Evaluation

Experiments were conducted on three clusters (32‑core/128 GB and 64‑core/256 GB) comparing OceanBase 4.1.0 (OB‑Cloud) with commercial distributed databases CloudDB‑A, CloudDB‑B, and DB‑A/DB‑B. Benchmarks (RO, RW, WO, Insert, Update) simulated Gaode’s workloads using Sysbench, scaling request rates up to 100 k TPS.

Results show that OB‑Cloud achieves higher throughput and lower average response time in both read‑intensive and write‑intensive scenarios, thanks to early lock release, OMS offline sync, and unit‑level request handling.

Storage compression tests reveal that OceanBase’s column‑store and micro‑block compression reduce storage size by 8.2× compared to DB‑A and 2.8× compared to DB‑B.

Availability tests under eight fault scenarios (process stop, node restart, network partition, etc.) measured Recovery Time Objective (RTO). OceanBase consistently recovered within 6 seconds, demonstrating rapid fault tolerance.

Conclusion

The study demonstrates that OceanBase’s unitization architecture provides high availability, high throughput, and efficient storage for large‑scale online map services. Future work includes migrating structured and vector data to OceanBase to further reduce costs and exploring its use for unstructured data workloads.