Scaling Property Services: StarRocks‑Powered Storage‑Compute Separation for 8000+ Communities

Every morning, the frontline team of Bifeng Service (碧服) conducts routine inspections across thousands of communities, while the backend system already aggregates tens of thousands of device statuses and user feedback. Serving more than 8,000 communities, the company faces a massive data lake of device monitoring, resident requests, and property operations data.

To handle this data flood while reducing costs and ensuring stable operation, the technical team introduced StarRocks with its real‑time processing and data acceleration capabilities, building an efficient and stable data engine that dramatically improved fault prediction, precise reporting, and real‑time monitoring.

OLAP Query Engine Evolution

Initially, data analysis relied on relational databases and ElasticSearch, suitable only for small‑scale, simple scenarios. As data volume grew, the company adopted Alibaba Cloud Hologres for real‑time warehousing, ClickHouse for federated queries, and in 2022 introduced the open‑source analytical database StarRocks. StarRocks’ multi‑table join capability boosted analysis efficiency and became the primary engine.

Challenges of the Monolithic Architecture

Node scaling difficulty: storage and compute were tightly coupled, requiring simultaneous scaling and causing performance impact during replica migration.

Low resource utilization: multiple independent clusters could not share resources, preventing elastic scaling.

Limited concurrency: peak‑time write‑heavy workloads slowed queries and degraded user experience.

Increasing SLA risk: growing data volume and complex OLAP queries caused latency spikes and higher failure risk.

To address these issues, the team adopted a storage‑compute separation architecture. Data is stored in OSS (object storage) with high reliability and scalability, while compute nodes can be independently scaled. An acceleration cache disk on compute nodes stores hot data, greatly improving query speed. From version 3.3 onward, lake‑warehouse integration was further enhanced.

Migration Implementation

The migration was treated as a high‑risk "data big migration". The team first audited existing cluster resource usage, then built a StarRocks cluster with 1 TB cache disks and multiple compute nodes. Over 200 business reports (1400+ tables, 1100+ integration tasks) were planned for migration.

To ensure a seamless switch with no user impact, a "dual‑write, gradual cut‑over" strategy was used. Data was written to both old and new clusters, nightly switches were performed after thorough validation, and fallback mechanisms were prepared.

Results and Business Value

Cost reduction : Hardware costs dropped as compute resources can be flexibly adjusted and storage moved to low‑cost OSS.

User experience improvement : Query P99 latency improved 8×, error rate reduced 30×, SLA reached 99.99%.

Performance gains : Real‑time dashboards and data services became 3‑5× faster.

Beyond the migration, the team continues to optimize parameters, data ingestion methods, table models, and query statements, while pursuing self‑built cluster capabilities and lake‑warehouse fusion using StarRocks’ integrated features.

Technical Selection and Planning

StarRocks, a widely adopted open‑source OLAP engine in China, offers high scalability, performance, and flexibility, making it suitable for the company’s rapid growth. Detailed migration planning and extensive testing were critical to the project’s success, with close collaboration among platform engineering, architects, data, and business system teams.

Future Optimizations

Post‑migration, the team will focus on continuous monitoring, parameter tuning, model redesign, and elastic resource allocation to further enhance read/write performance and stability.