The Evolution and Practice of Multi-Write Multi-Read Shared Storage Clusters – UXDB SRAC Whitepaper

Market Landscape and Strategic Value

The paper notes that the global DBMS market is projected to reach $1.61 trillion by 2026, with relational databases holding about 62% of the share. High‑end cluster databases that support multi‑write multi‑read shared storage form a niche with the highest technical barriers and strategic importance, serving critical workloads in finance, telecom billing, and energy scheduling where continuous availability and strong data consistency are mandatory.

In China, foreign products such as Oracle RAC have long dominated core information infrastructure. Driven by the ongoing “Xinchuang” policy, domestic databases are increasingly replacing foreign solutions in core transaction systems across finance, energy, and government sectors.

Evolution of Shared‑Storage Cluster Technology

The historical roots trace back to the 1986 GAMMA database project at the University of Wisconsin‑Madison, which explored parallel processing on shared‑storage systems. In 2001, Oracle’s seminal paper “Cache Fusion: Extending Shared‑Disk Clusters with Shared Caches” introduced cache‑fusion, reducing data‑block transfer latency from milliseconds (disk I/O) to microseconds (network), achieving 2–3 orders of magnitude performance gains. Subsequent products such as IBM Db2 pureScale built on these ideas.

Chinese academic groups—including East China Normal University, Tsinghua University, and Renmin University—have contributed research on cache‑fusion optimization and deterministic concurrency control. Industry players such as Youxuan, Alibaba Cloud, Huawei, and openGauss have also achieved breakthroughs.

UXDB SRAC: Core Innovations of a Fully Symmetric Multi‑Write Multi‑Read Architecture

UXDB SRAC adopts a fully symmetric multi‑write multi‑read shared‑storage design where every node is equal, capable of handling both read and write requests on the full data set. Its architecture consists of three major subsystems:

Cache Fusion – Merges the shared‑memory buffers of all nodes into a single global cache, allowing any node’s data‑page modifications to be perceived and fetched by other nodes with ultra‑low latency while preserving consistency.

Global Transaction Concurrency Control – Extends ACID guarantees to the cluster level via a global transaction manager, BOC broadcast commit, MVCC, and a Global Lock Service (GES), ensuring strong cross‑node transaction consistency.

Shared‑Storage File System – A self‑developed stack from block devices to file interfaces that enables concurrent reads and writes to the same set of files across nodes, with strong metadata consistency.

For high availability, UXDB SRAC implements a three‑layer protection scheme (heartbeat detection, fault arbitration, online recovery) achieving RPO = 0 (zero data loss) and RTO < 10 seconds within the same city.

Authoritative Certifications and Benchmark Results

In June 2026, UXDB SRAC passed the China Academy of Information and Communications Technology’s “Database Basic Capability on Shared‑Storage Architecture” assessment, meeting all 39 technical criteria. It also received a national scientific‑technology achievement award from the Chinese Institute of Electronics, praised for its technical complexity, innovation, and internationally advanced key metrics.

Benchmark case studies demonstrate:

In a national research institute’s simulation platform, sustained writes approaching 100 M operations per second with RTO < 5 seconds and automatic fault self‑healing.

In energy‑dispatch scenarios, continuous ingestion of over 500 k data samples per second.

Readiness for core financial trading, telecom support, and public‑safety applications.

Industry Outlook

The paper asserts that multi‑write multi‑read shared‑storage clusters remain the most technically challenging DB architecture, with only a handful of teams worldwide having fully solved the core problems and productized them. UXDB SRAC’s deployment marks a substantive breakthrough for domestic databases in this domain.

Future work will focus on optimizing resource‑migration algorithms, enhancing backup‑recovery capabilities, and improving cluster observability. Expansion plans target deeper penetration in finance, telecom, and energy, accelerated adoption in government “Xinchuang” initiatives, and exploration of overseas promotion through international cooperation frameworks.

The authors conclude by honoring database pioneers such as Professor Michael Stonebraker and Oracle RAC engineer Roger Bamford, emphasizing that database product success is a marathon and that Youxuan Database is committed to long‑term, collaborative ecosystem development.