Returning to Nanda Tongyong After 15 Years: A Dual Reunion of Time and Technology

Prologue: Meeting Old Colleagues

Walking into the conference hall, the author feels a mix of professional curiosity and personal nostalgia when familiar faces appear.

1. From Intern to 2011 Nanda Tongyong

In 2011, as a graduate student, the author spent nearly a year as a technical‑support intern, learning to deploy and test GBase, write technical PPTs, and troubleshoot production issues by examining logs and monitoring data. These experiences formed the core methodology used throughout a decade‑long DBA career.

2. Nanda Tongyong and GBase: A Notable Chapter in Domestic Databases

Nanda Tongyong, founded in 2004, has been a leading Chinese database vendor for over twenty years, holding a key position in the domestic “information technology application innovation” market.

Three product lines cover analysis, transaction, and multi‑model scenarios, forming the complete GBase family.

3. GBase 8a: From Analytic MPP to an Integrated Lakehouse Platform

Product Positioning

GBase 8a is the flagship column‑store MPP data‑warehouse product, now expanded into several forms:

8a Single‑Node – up to 50 TB for high‑performance analytics.

8a Cluster (MPP) – petabyte‑scale analytics with lake‑warehouse integration.

8a Cloud Data Warehouse (GCDW) – compute‑storage separation, cloud‑ready, cost‑effective.

GBase HD – Hadoop‑compatible lake‑warehouse.

GBase 8a DataAgent – data + AI intelligent agent for end‑to‑end analytics.

Key Technical Directions

1. Compute‑Storage Separation (GCDW) – Metadata and data are fully decoupled, components are stateless, and compute resources can elastically scale. Multi‑tenant isolation is supported via independent warehouses, and open‑format data (Parquet, ORC) can be accessed directly from HDFS/S3.

Cost advantages: hardware procurement reduced by 50‑90 %, storage cost cut by 70‑90 %, and scaling efficiency improved by 70‑90 % (seconds‑level compute expansion without resharding).

2. Column‑Store Transactions for OLAP – GBase 8a adds transaction capability to a column store using GTM (transaction ID generation), MVCC, and WAL logs, providing row‑level locks and thousands of concurrent transactions while preserving analytic performance.

3. Dual‑Active Cluster (GVR) – Supports same‑city dual‑active, multi‑center high‑availability, read/write separation, and heterogeneous hardware replacement via graphical configuration.

4. Data + AI (DataAgent) – Combines lake‑warehouse architecture, semantic layer, natural‑language query, and full‑stack intelligent operations. Three parallel business lines:

DataAgent (business) – natural‑language query, intent understanding, model interpretation.

GDS (development) – SQL assistance, quality management, query modeling.

GDOM + DBClaw (operations) – automated monitoring, fault analysis, root‑cause diagnosis, resource scheduling.

The semantic layer eliminates AI hallucination by unifying business, data, and technical ontologies.

GBase’s AI vision shifts from “machine‑assisted human (CAX)” to “human‑assisted machine (HAX)”, turning data consumers into agents.

4. GBase 8c: Multi‑Model, AI‑Native Database

GBase 8c represents the modern evolution of transactional databases and a systematic response to AI challenges.

Multi‑Model Unification

Row Store (OLTP) – high‑concurrency point queries.

Column Store (OLAP) – column‑wise I/O, high compression, complex aggregates.

HTAP (Hybrid) – optimizer auto‑routes TP to row store, AP to column store, with asynchronous delta sync (“one data, two services”).

Deployment modes include primary‑backup, distributed, and compute‑storage separation, adaptable to various scales.

Compatibility is achieved via the DBCOMPATIBILITY parameter, supporting Oracle, MySQL, PostgreSQL, and SQLServer dialects.

AI‑Era Technical Responses

GBase 8c identifies five AI‑era challenges: exploding data volume, elastic scaling, high‑concurrency low‑latency, vector retrieval, and data flow efficiency.

1. Native Vector Storage – Enables SQL joins between vector and scalar data, using bitmap filtering and ANN algorithms for RAG scenarios.

2. Compute‑Storage Separation – Stateless compute nodes, WAL‑based stream processing, and distributed storage achieve scale‑to‑zero and seconds‑level recovery.

3. Data Branching (Copy‑on‑Write) – Branch creation records only WAL position; new branches share historical pages, copying on write, enabling rapid experiment branching for AI model training.

4. LTAP (Lake‑house Transaction/Analytics Processing) – Replaces the traditional ETL pipeline with real‑time mirroring from OLTP (GBase 8c) to lake‑house (GBase 8a), providing fresh data for analytics and AI agents.

5. GBase 8s: Enterprise‑Grade Centralized Transactional Database

GBase 8s, the longest‑standing product line, is widely deployed in finance, telecom, and government.

Key capabilities:

High reliability – RPO = 0, RTO < 3 s, transparent failover.

High security – EAL4, multiple certifications.

High concurrency – coroutine + lock‑free kernel, ten‑thousand‑level concurrent sessions.

Large capacity – single table up to 16 TB, single node hundreds of TB.

Technical Highlights

1. In‑Memory MVCC – Eliminates I/O bottlenecks and disk bloat; version data stays in memory and is reclaimed by aggressive GC.

2. HK‑Tree Index – Designed for low‑cardinality data, offering high‑density storage, sequential read‑back, GROUP + SORTED properties, and HTAP support. Benchmarks on 360 M rows show query times of 0.x s versus 2‑8 s for B‑Tree, and aggregation speeds orders of magnitude faster.

3. High Concurrency – TPC‑C 1000‑warehouse tests show performance degradation only after 2000 concurrent threads; peak TpmC exceeds 1 M, and even at 10 k concurrency performance remains around 800 k TpmC without collapse.

4. TAC (Transaction Augmented Continuity) – Guarantees session, transaction, and SQL continuity across DQL/DML/DDL, with transparent primary‑backup switching.

5. Compatibility and Migration Toolchain – Near‑full Oracle compatibility, >90 % compatibility with MySQL, PostgreSQL, DB2, plus MTK migration tools and RTSync real‑time sync.

6. DBA Perspective on the Conference

The author notes that GBase’s DataAgent and DBClaw align with a vision of a semantic layer bridging AI and data, turning repetitive DBA tasks into agent‑driven operations.

Real‑world substitution cases: Shenzhen Metro Phase 4 replaced Oracle with GBase 8s (220 nodes, 1.3 B rows, 13 k concurrent sessions) achieving 60 % cost reduction; a trust replaced MySQL, migrating 936 k tables and tripling TPS.

Technical gaps with international products are narrowing; innovations such as compute‑storage separation, native vector retrieval, copy‑on‑write branching, and in‑memory MVCC place domestic databases ahead in certain dimensions.

Conclusion

The article serves as a delayed thank‑you to the 2011 internship and a careful observation of Nanda Tongyong’s 22‑year technical accumulation, emphasizing that the database field demands long‑term, hands‑on perseverance and that seasoned engineers embody a rare depth of experience in the internet era.