CynosDB: Tencent Cloud's Next-Generation Enterprise Distributed Cloud Database

CynosDB (also known as NewCDB) is a next-generation enterprise-level distributed cloud database jointly developed by Tencent Cloud and TEG Infrastructure Department Database Technology Team for the cloud computing 2.0 era. Compatible with the open-source MySQL ecosystem, it redefines MySQL distributed architecture to achieve commercial-grade database performance in single-instance extreme performance, transaction response latency, storage scalability, and availability. It represents extreme software optimization for general-purpose hardware, offering very high cost-performance ratio.

The name CynosDB originates from Latin "Cynosura," the Greek mythological name for Polaris, meaning focus or something that attracts attention. CDB is the product abbreviation for Tencent Cloud Database, and NewCDB is built upon ten years of CDB technology and product accumulation.

With the advent of cloud computing 2.0, more internet businesses require higher performance, stability, and lower costs from database services. Traditional industries undergoing digital transformation need cloud databases that can seamlessly replace traditional products. Simply lifting and shifting Oracle or SQL Server to the cloud cannot meet scalability and reliability requirements.

CynosDB参照了类AWS Aurora路线, providing a new MySQL service with next-generation architecture. The main technical advantages include: Cloud-native design based on Tencent's mature components (TXSQL and TXStore); Compute-storage separation; Distributed storage providing high availability, reliability, and strong consistency; Computable storage where logs become the database and storage completes log-to-page conversion independently; Lightweight inter-node replication with low overhead and low latency.

The compute engine uses Tencent MySQL (TXSQL) supporting MySQL 5.6/5.7 with over 100 custom patches for performance, enterprise features, and stability. The storage engine uses Tencent Storage (TXStore) as the unified storage platform. Through decoupling, storage is provided independently via storage pooling to solve storage expansion, write performance, and resource utilization issues.

The system uses a minimalist IO approach, transmitting transaction logs with storage completing log-to-record conversion. It leverages RDMA hardware advantages with flexible server-initiated and client-initiated data transfer modes. Storage is shared among multiple compute nodes, enabling rapid compute node expansion with zero-copy data operations.

TXStore achieves 99.9999999% (9 nines) data reliability through multi-version second-level snapshot technology combined with CDP (Continuous Data Protection). Using software-hardware integration, it provides millions of IOPS per client and sub-millisecond response time, utilizing NVME/Optane storage media and RDMA networks. AI deep learning enables disk failure prediction, IO prefetch, and caching capabilities.