2022 China Database Industry Report: Emerging Hardware and Architectural Innovations

The September 2022 China Database Industry Analysis report shows that database systems are rapidly evolving alongside processor, memory, storage, and network hardware, leveraging multi‑core CPUs, heterogeneous GPUs/TPUs/DPU, programmable FPGAs, CXL‑DDR5, persistent memory (PMem), storage‑class memory (SCM), NVMe+SSD+NVMe‑oF, and RDMA‑enabled software‑defined storage to meet massive data and real‑time compute demands.

X‑SSD is a transaction‑log‑optimized SSD architecture that provides a low‑latency byte‑addressable fast path for log writes while retaining a traditional SSD block interface for regular workloads, allowing seamless data transfer between the two paths.

Gacc0 is a GPU‑accelerated OLTP engine that splits execution between CPU and GPU, grouping similar transactions for vectorized GPU processing and using CPU memory to extend beyond GPU memory limits, achieving up to a 6× performance boost in TPC‑C benchmarks.

Halo introduces a hybrid PMem‑DRAM persistent hash index to improve write performance and durability.

Sherman is a distributed B+‑tree index built on RDMA‑enabled servers; it reduces round‑trips by merging RDMA commands, employs a hierarchical lock using on‑card memory, and adopts a two‑level versioning scheme to cut write amplification, delivering an order‑of‑magnitude throughput and latency improvement over state‑of‑the‑art designs.

Zen is a log‑less, high‑throughput OLTP engine for NVM that uses a hybrid table (HTable) composed of an NVM tuple heap, a DRAM meta‑cache, and per‑thread NVM tuple managers, storing metadata and coarse‑grained allocation structures in NVM while keeping indexes and transaction‑private data in DRAM.

TSCache is a flash‑based distributed time‑series cache that offers a simple key‑range interface for storing and retrieving time‑series data without acting as a middle layer to backend TSDBs, thus serving multiple time‑series databases uniformly.

Crystal is a “smart” storage middleware that decouples databases from raw storage, operating as a mini‑DBMS or cache management system with a CMS component on compute nodes and database‑specific connectors that push predicates to underlying data sources such as CSV or Parquet.

SPADE is a GPU‑accelerated spatial database engine that supports rich spatial queries using a canvas data model and GPU‑friendly algebra, leveraging graphics pipeline operations for portability across GPU hardware.

Exadata X9M integrates cloud‑ready architecture, scalable high‑performance database servers, intelligent storage with PCIe flash and persistent memory, and RoCE networking, delivering superior performance, cost efficiency, and availability for OLTP, DW, analytics, IoT, gaming, and other workloads.

openGauss introduces kernel innovations for many‑core and large‑memory servers, optimizing concurrency control algorithms, kernel data structures, and data access paths, achieving 1.5 million tpmC on dual‑Kunpeng 128‑core platforms and incorporating a Memory‑Optimized Table (MOT) engine for higher transactional performance.

The report also surveys integrated database appliances such as DAMENG PAI, PBData‑V, zData, QData, and TData, each combining high‑performance compute, flash storage, RDMA networking, and multi‑tenant, high‑availability designs to serve diverse OLTP, OLAP, and hybrid workloads.