How to Optimize Chinese Domestic Databases: Practical DBA Tips Beyond Benchmarks

DBAs who have worked with Oracle are increasingly required to support Chinese domestic database products, and the article examines which of their existing skills remain useful and which need adjustment.

The first recommendation is to avoid copying Oracle benchmark‑driven parameter changes directly. Domestic databases often advertise extremely high TPMC (transactions per minute per core) numbers—sometimes exceeding one million on a dual‑CPU server—making raw benchmark figures misleading. The author notes that a TPMC of 250 k was sufficient for a core system, while many vendors claim 400 k or more, yet such claims may not translate to real‑world stability.

Second, traditional Oracle metrics like buffer‑hit rate are far less indicative for these systems. Most domestic products derive from open‑source engines such as PostgreSQL or MySQL, so their buffering architecture combines OS and database caches. Consequently, focusing on OS‑level settings—dirty‑page flush policies, virtual memory allocation, NUMA parameters—and minimizing physical I/O latency yields more noticeable performance gains than simply enlarging the database buffer pool.

Third, SQL execution efficiency is critical because the cost‑based optimizer (CBO) in many domestic databases lags behind Oracle. Building a performance baseline for execution plans and regularly comparing core SQL statements against that baseline helps detect plan degradation early and prevents severe slowdowns.

Fourth, high‑concurrency workloads often expose issues not seen on Oracle. Problems such as hot‑block conflicts and spin‑lock contention can cripple throughput. The author shares a case where adjusting the data‑block fill‑rate and switching a range‑partitioned table to HASH partitioning raised TPMC by 40 %. Additionally, using tools like perf to identify spin‑lock wait hotspots and tuning related kernel parameters can resolve bottlenecks.

Overall, optimization for domestic databases is still an exploratory field with limited vendor guidance. As adoption grows, practical experience and community‑driven knowledge will become essential, and vendors are encouraged to invest beyond TPMC benchmarks toward deeper performance‑tuning documentation.