GeaGraph: Large-Scale Graph Computing System Wins World Internet Conference Award

On September 26, the 2021 World Internet Conference Wuzhen Summit opened.

Fourteen world‑leading internet technology achievements were announced that day, and the Ant Group together with Tsinghua University’s independently developed "Large‑Scale Graph Computing System GeaGraph" project was selected for an award.

(Image: CCTV2 live broadcast of the award announcement)

High‑performance graph computing is considered a key frontier of global cutting‑edge technology, with broad application prospects in telecommunications, healthcare, manufacturing, energy and other fields.

The Ant Group and Tsinghua University jointly developed the large‑scale graph computing system GeaGraph, which can perform real‑time queries on trillion‑edge graphs and ranked first in international standard graph database tests, achieving internationally leading scale and performance.

Chinese Academy of Engineering academician and Tsinghua University professor Zheng Weimin said: “GeaGraph gives us a good example: industry‑academia‑research collaboration, where universities and leading tech companies jointly overcome technical difficulties and scale the solution.”

(Image: Ant Group Chairman and CEO Jing Xiandong and the Ant Graph Computing team at the award ceremony)

Strengthening industry‑academia‑research collaboration is an effective model for accelerating the upgrade of China's independent basic software

On September 26, I was delighted to see the Ant Group and Tsinghua University’s joint research project "Large‑Scale Graph Computing System GeaGraph" listed among the World Internet Conference leading technology awards, representing an internationally leading achievement in a branch of system software and providing a successful example for domestic basic software development.

Graph models abstract entities and relationships as nodes and edges on a graph, enabling deeper relational analysis than traditional relational models and offering wide‑range applications in financial anti‑fraud, anti‑money‑laundering, internet search, intelligent manufacturing, and energy internet.

Since around 2010, Tsinghua University’s High‑Performance Computing Lab has been researching graph computing technologies, producing systems such as the Gemini graph engine in 2016, which outperformed common open‑source engines like GraphX by about 100×. In 2016, graduates founded Fermat Technology Co., which later developed the internationally leading graph database product TuGraph, supporting full‑transaction graph operations and achieving the top score in the LDBC certification test in 2020.

Ant Group, with the largest user base and peak transaction volume in China and globally, has rich graph‑computing scenarios such as anti‑fraud and anti‑cash‑out for Alipay. Starting in 2015, Ant independently developed distributed graph databases and streaming graph computing systems, achieving good internal results.

In 2020, Ant integrated its own technologies with those from Tsinghua and Fermat to upgrade into a complete graph computing system called GeaGraph. This system combines the strengths of all parties and, without exaggeration, reaches world‑leading levels in functionality, throughput, and response time.

Nevertheless, we should not become complacent. The global graph computing field is still in its early stages; graph query languages lack a solid theoretical foundation comparable to relational algebra, many graph databases have poor write performance, and most cannot pass international standard tests. Techniques such as materialized views that could dramatically improve complex query performance are largely absent in graph databases. More proactive industry‑academia‑research collaboration is needed to further enhance China’s leading advantage in this niche of system software.

We also hope that successful cases like GeaGraph can be replicated at scale, greatly accelerating the development of advanced basic software in China.

I have always emphasized that domestic basic software should not merely be low‑level replacements or re‑branded open‑source; we must learn to "build advanced system software from scratch." GeaGraph provides a clear example of how industry‑academia‑research collaboration can overcome technical challenges and achieve large‑scale application.

We look forward to more Chinese universities and technology companies joining this model. This full‑chain practice of "industry‑academia‑research" collaboration is one of the secrets of Silicon Valley’s success, and if smoothly implemented and widely copied in China, it will significantly speed up solving the bottleneck problems in our basic software field and gradually build advanced foundational software developed domestically.