How REDgraph Supercharges Query Performance for Massive Social Networks

Introduction

Xiaohongshu, a community‑centric product with massive social network data, faced challenges in social‑scenario applications and distributed parallel query implementations. To address these, they developed REDgraph, a graph database designed for ultra‑large‑scale social networks, dramatically improving query efficiency and performance.

Presentation Outline

The talk is divided into five parts: background introduction, analysis of the original architecture problems, distributed parallel query implementation, summary and outlook, and a Q&A session.

Background: Graph Database Overview

Graph databases belong to the NoSQL family. Compared with other NoSQL types—key‑value, wide‑column, document, and time‑series—their data association degree and query complexity increase from KV to graph. While KV, wide‑column, and document focus on richness within a single record, graph databases specialize in handling relationships, making them ideal for scenarios that require deep link or multi‑dimensional relationship mining.

Comparison with Relational Databases

Consider a typical social‑network schema with user, friend, like, and note tables. Retrieving the notes liked by a user's friends in a relational database requires a lengthy SQL statement with multiple JOIN operations, consuming significant CPU, memory, and I/O resources. Even with carefully designed indexes, the cost remains high and maintenance intensive.

Using a graph database simplifies the process. By modeling two vertex types (users and notes) and two edge types (friendship and likes), the data forms a clear network structure. A Gremlin query can retrieve the desired information in just four lines:

g.V().has('name','Tom')
  .out('friend')
  .out('like')
  .values('content')

This query is concise, readable, and directly reflects the underlying graph topology.

Key Advantages of Graph Databases

Graph databases store vertices and edges as first‑class citizens, enabling extremely efficient adjacency and relationship traversals. Even as data volume grows, query latency remains stable, avoiding the performance degradation typical of relational joins.

Source Note

Excerpted from the e‑book AI for Data: Intelligent Data Processing and Analysis Practice . Scan the QR code below to join the community and obtain the e‑book.