Exploring Tencent Cloud Native Database TDSQL-C Architecture and Practices

From May 20‑22, 2021, the 13th China System Architect Conference (SACC2021) was held online, where Tencent Cloud database expert Shang Bo delivered a talk titled "Exploring Tencent Cloud Native Database Architecture and Practices".

Part 1: Background and Architecture of Cloud‑Native Database TDSQL‑C With rapid growth of internet services, data volumes have exploded, demanding higher storage capacity, reliability, availability, and scalability such as multi‑hundred‑TB storage, transparent scaling, multi‑replica disaster recovery, and sub‑second failover. Traditional databases struggle to keep up, while distributed architectures offer clear advantages. TDSQL‑C adopts a novel compute‑storage‑separated distributed design, fully compatible with MySQL and PostgreSQL, and implements shared cloud storage, log transport, data sharding, compute‑storage, and physical replication, achieving PB‑level storage, over 1 million QPS, second‑level RTO, millisecond‑level standby latency, snapshot backup, parallel rollback, and flexible cluster or serverless deployment, reducing total cost by up to 90% for enterprise users.

Part 2: Key Scenario Breakthroughs in Practice In terms of availability, TDSQL‑C Serverless provides ultra‑fast start‑stop and elastic scaling, automatically adjusting instances based on load, supporting seamless scaling, pay‑per‑use billing by second, and flexible CPU/memory configuration. For performance, the system introduces a shared‑memory buffer pool and a separate hot‑data secondary cache, leveraging non‑volatile storage to boost hot‑data access, achieving over 100% performance improvement for IO‑bound workloads as data volume grows. Regarding reliability and availability, each storage shard can independently back up based on backup points while maintaining global snapshot consistency; parallel backup and rollback across shards can reach GB/s speeds.

Part 3: Future Exploration of TDSQL‑C Future work includes adding an independent Log Store to accelerate log response and overall throughput, delivering extreme write performance, cross‑region read services, and cross‑region disaster recovery for finance‑grade reliability. Deep kernel optimizations will break conventional log‑first designs, enabling fast page eviction, partitioned concurrent reads/writes, and remote page writes, further enhancing high‑concurrency performance and stability. The compute‑storage‑separated architecture will continue to adopt log‑as‑data concepts, multi‑dataset partitioning, multi‑log streams, peer‑to‑peer writes, and global transactions across data sets, maximizing overall read/write throughput and system availability.

TDSQL‑C is dedicated to helping enterprises reduce costs and increase efficiency while delivering maximum commercial value; more detailed product information is available on the Tencent Cloud website.