Eye or the Tiger: Benchmarking Cassandra vs. TimescaleDB for Time-Series Data

This article compares Cassandra and TimescaleDB for handling time-series data, focusing on their scaling patterns, data model complexity, insert rates, read rates, and read-throughput. The results show that five TimescaleDB nodes outperform a 30-node Cassandra cluster with higher inserts, up to 5,800x faster queries, 10% of the cost, a much more flexible data model, and full SQL. While Cassandra's clustered wide rows perform well for querying data for a single key, it quickly degrades for complex queries involving multiple rollups across many rows. Additionally, while Cassandra makes it easy to add nodes to increase write throughput, it turns out you often just don’t need to do that for TimescaleDB. With 10–15x the write throughput of Cassandra, a single TimescaleDB node with a couple of replicas for high availability is more than adequate for dealing with workloads requiring a 30+ node fleet of Cassandra instances to handle.