Case Study: IBM Graph achieves major performance gains for JanusGraph with ScyllaDB

Scylla Shines in IBM’s Performance Tests for JanusGraph

IBM Graph, IBM’s enterprise graph DBMS service built on JanusGraph for real-time analytics and recommendations, needed a higher-performance storage backend than the Cassandra and HBase options it had been using. To evaluate alternatives, IBM’s Open Tech and Performance teams tested ScyllaDB as a storage backend for JanusGraph in their 3-node clusters to see whether ScyllaDB could meet demanding write, read/write edge insert, and complex query workloads.

ScyllaDB outperformed both HBase and Cassandra across all tests: vertex inserts showed roughly 35% higher throughput than HBase and nearly 3× Cassandra; edge inserts were about 160% better than HBase and over 4× Cassandra; and query throughput was ~72% better than Cassandra and nearly 150% better than HBase. ScyllaDB also delivered much lower latency (about half Cassandra’s and one-third HBase’s at high load), simpler clustering, effective self-tuning and load balancing, full resource utilization, and compatibility with existing Cassandra clients—making ScyllaDB the clear choice for IBM Graph’s JanusGraph storage backend.

IBM Graph

Chin Huang

Software Engineer