Why Netflix skipped native graph databases (Neo4j, Neptune) for its Real-Time Distributed Graph and built on KVDAL — a key-value abstraction over Apache Cassandra — instead. Jordan and Riley cover the record/item mapping that makes a key-value store hold a graph (nodes as property records, edges as adjacency lists), the per-type namespace model for independent scaling, and the scale: 8 billion nodes, 150+ billion edges, 2M reads/sec and 6M writes/sec at single-digit-millisecond latency across 2,400 EC2 instances.

Source: How and Why Netflix Built a Real-Time Distributed Graph, Part 2: Building a Scalable Storage Layer — Netflix Tech Blog, Nov 14 2025 —  ...