Evolution and Architecture of Tencent Cloud Block Service (CBS): From CBS1.0 to CBS3.0

What is Cloud Disk? Cloud Disk (Cloud Block Service, CBS) is a block storage service on Tencent Cloud that provides traditional disk capabilities—file systems, snapshots, and high‑availability—while leveraging cloud resources.

How CBS evolved CBS started as CBS1.0, which stitched together three existing distributed systems (TFS, TSSD, CKV) and exposed an iSCSI front‑end. The heavy dependency caused long I/O and operation chains, prompting a redesign to CBS2.0 that simplified and integrated the three systems at the code level.

CBS2.0 Architecture The front‑end consists of a Client (iSCSI initiator), a Proxy (iSCSI target), and a Master node for coordination. The back‑end is a distributed storage cluster with an Access module, Chunk storage, and its own Master. This two‑cluster design improved manageability but still suffered from cost and latency issues.

Operational Status CBS2.0 has been running safely for years, serving hundreds of thousands of commercial customers, managing hundreds of thousands of disks, and storing data at the hundred‑petabyte scale with 99.999999% reliability.

Key Challenges The main problems were high operational cost and latency in high‑performance scenarios, where the multi‑layer network path added tens of microseconds comparable to SSD access latency.

Simplifying to CBS3.0 The solution was to remove the access layer, creating a two‑layer architecture: a Client (Driver) on the host and a Chunk module for storage, overseen by a Master node. This reduces cost by 46% and supports HDD, HDD+SSD, and SSD disk types.

Technical Challenges of the Two‑Layer Design The critical issues are data organization, data routing, and routing synchronization. CBS introduces virtual partitions (Partition) that map fixed‑size Blocks to logical Partitions, balancing routing table size and storage efficiency.

Data Routing Requests carry diskId, blockId, and snapId; a consistent‑hash algorithm maps these to a Partition, whose physical location is configured during cluster initialization.

Lazy Routing Synchronization To avoid overwhelming the Master with route updates, CBS adopts a “lazy” approach: the Master pushes routes only to Chunk nodes, Chunk stores them locally, and Drivers request updates only when needed. This lazy‑routing algorithm reduces synchronization overhead.

CBS3.0 Benefits After deployment, CBS3.0 cut ordinary cloud‑disk costs by 46%, provided a unified platform for all disk types, and delivered higher performance than comparable open‑source solutions such as Ceph.

Industry Comparison While both CBS and Ceph use a two‑layer architecture, CBS offers superior performance, finer‑grained operations, and stronger data safety guarantees, whereas Ceph adds features like object storage and erasure coding.

Conclusion The evolution of CBS demonstrates a “less is more” philosophy: by continuously simplifying the system architecture, Tencent achieved higher performance, reliability, and cost efficiency while delivering the essential functionality users need.