REDtao: A High-Performance Graph Storage System for Social Graph Data

REDtao is a self-developed graph storage system designed to address the challenges of managing super-large-scale social graph data for Xiaohongshu (Little Red Book). It leverages Facebook's Tao architecture while incorporating optimizations tailored to the platform's specific needs, such as high read/write throughput, data consistency, and scalability. The system decouples caching and persistent storage layers, enabling flexible scaling and reducing reliance on MySQL. Key features include a multi-layer hash-based data structure for efficient point-edge queries, distributed caching with leader-follower architecture, and cross-cloud multi-availability support. REDtao achieved 90%+ cache hit rates, reduced MySQL CPU usage by 70%, and supported 150W QPS on a single 16-core machine. The migration from MySQL to REDtao was completed without service interruption, handling a 250%+ increase in DAU-driven requests.

For code content, the system's data structure is implemented as a three-layer nested HashTable. Each layer handles different aspects of graph data: the first layer maps from_id to REDtaoGraph, the second layer tracks AssocType metadata, and the third layer stores edge details. This design allows efficient memory management and high query performance. The REDtaoQueue mechanism ensures ordered processing of read/write requests for the same edge, preventing data inconsistencies.