DataStax’s Apache Cassandra has emerged as the top performer among leading NoSQL databases, both in throughput and latency.
Commissioned by DataStax and conducted by consulting company End Point, the tests showed that Apache Cassandra 2.1.0 has significantly outperformed Couchbase 3.0, MongoDB 3.0 (with the Wired Tiger storage engine), and HBase 0.98, in throughput and latency.
End Point has tested the databases on Amazon Web Services EC2 instances, using the open-source specification and programme suite Yahoo! Cloud Serving Benchmark (YCSB).
The consultancy group scaled each database from 1 to 32 nodes for a variety of tests, including load, insert heavy, read intensive, analytic, and other typical transactional workloads.
The report, Benchmarking Top NoSQL Databases, focused on workloads, data volumes, and conditions most commonly found in production environments.
End Point used a multi-node, scale out implementation with data volumes that exceeded the RAM capacity on each machine with no possibility for data loss during load/write operations.
The tests used the i2.xlarge class of instances (30.5GB RAM, four CPU cores, and a single volume of 800GB of SSD local storage) for the database nodes. The c3.xlarge class of instances (7.5GB, 4 CPU cores) were used as client nodes.
Small, RAM-only data volumes, and single node deployments were ignored.
End Point originally performed tests in February 2013. However, with performance enhancements made to each database, DataStax commissioned the consultant to do the tests once again by including Couchbase to its list.
According to End Point, results of the latest tests seconded those of the 2013 analysis.
End Point CTO Jon Jensen said: "NoSQL databases have become a standard part of the toolset for high-throughput, horizontally scalable applications with large data sets, and Apache Cassandra proved to be the top performer throughout this study.
"While we always recommend that anyone assessing a database’s performance test the engine under specific use case and deployment conditions intended for a particular production application, general competitive benchmarks like this one can be very useful to those evaluating NoSQL databases."