Graph Database Storage Technologies and Practices: Concepts, Core Goals, Technical Solutions, and Galaxybase Case Study

The presentation begins with an overview of graph databases, describing the need for them in scenarios where massive, highly connected data arises—such as social networks, finance, retail, telecommunications, power systems, government, and manufacturing—and defining what a graph database is.

It then identifies the core storage objectives of graph databases, emphasizing the importance of deep‑link (multi‑hop) queries, massive data volumes, and low‑latency real‑time analysis, and argues that traditional relational databases cannot meet these requirements.

The article explains that the fundamental operation for association analysis is neighbor iteration, and that achieving index‑free adjacency—where a vertex and its adjacent edges are stored together—reduces the time complexity of each iteration to O(1), independent of the overall graph size.

Three storage schemes are examined:

Array‑based storage: points and edges are stored sequentially in arrays, offering fast sequential reads but suffering from variable‑length challenges and slower writes.

Linked‑list storage: uses fixed‑length IDs and offsets to avoid variable‑length issues, but incurs many random disk reads, requiring effective caching for acceptable performance.

LSM‑tree storage: leverages a multi‑level, log‑structured merge tree where edges are keyed by source vertex, enabling index‑free adjacency while providing high write throughput; however, read latency can increase due to layer traversal and compaction overhead.

Optimization trade‑offs are summarized: arrays give fast reads, LSM trees give fast writes, and linked lists offer flexibility but neither excels in speed.

The article then showcases Galaxybase, a native distributed graph database, highlighting its high performance, scalability to trillion‑scale graphs, real‑time analytics, efficient compression, and compatibility with both open‑source and domestic hardware ecosystems.

Performance benchmarks are presented, including a partnership with Sun Yat‑sen University that processed 5 × 10¹³ transaction records on 50 machines, achieving six‑hop queries in 6.7 seconds and surpassing previous world records. Additional results from the LDBC‑SNB benchmark demonstrate up to 72× higher throughput compared with prior state‑of‑the‑art systems.

Galaxybase also offers a rich set of graph algorithms (traversal, shortest path, centrality, community detection, similarity, subgraph matching) and integrates with cloud ecosystems such as Tencent Cloud, Baidu Cloud, and AWS, with deployments in finance, energy, education, government, and internet sectors.

The session concludes with a thank‑you note and a call for audience engagement.