Performance Evaluation: ScyllaDB as a Database Backend for JanusGraph

In their performance evaluation presented at the ScyllaDB Summit 2017, IBM engineers Ted Chang and Chin Huang analyzed ScyllaDB as a potential backend for JanusGraph, comparing it to traditional backends Cassandra and HBase. They conducted tests on three scenarios: vertex insertion (write operations), edge insertion (a mix of read and write operations), and graph traversal (read operations), using a 3-node cluster setup for each database. The results showed ScyllaDB excelling in throughput, with 272K vertices and 132K edges inserted per second, outperforming Cassandra and HBase significantly. ScyllaDB also demonstrated efficient CPU utilization, saturating at high levels compared to the other two databases, which showed lower CPU usage. In graph traversal tests, ScyllaDB exhibited superior transaction throughput while maintaining lower latency, reinforcing its performance advantage. The engineers concluded that ScyllaDB's design to maximize CPU usage under heavy loads aligns with its high performance, establishing it as a leading choice for graph database backends.