PLEASANTON, Calif.--(BUSINESS WIRE)--United Software Associates, Inc. today announced the results of its comprehensive independent database comparison, measuring performance across multiple dimensions. The research, based on the Yahoo! Cloud Serving Benchmark (YCSB), determined MongoDB overwhelmingly outperformed key value stores, in terms of throughput and latency, across a number of configurations.
The full performance benchmark study is available for download.
“In this research, our goal was to evaluate the performance of MongoDB and NoSQL databases, for reads and writes, across a range of durability scenarios, putting each product on a level-playing field. Our results show that in every scenario, MongoDB with WiredTiger clearly outperforms its competition,” said Sam Bhat, CEO of United Software Associates. “We were surprised to find that, even with its more extensive feature set, MongoDB outperformed key value stores at what they do best. However, as YCSB only tests a small set of the requirements necessary for any application, organizations should carefully test all their requirements to make smart choices about their database technology. We have posted the tests on GitHub so others can reproduce our findings.”
It is important for enterprises to determine the business values of advancements in technology. United Software Associates has over 16 years of technology adoption and value engineering experience to help customers understand the true cost and benefits of using many emerging technologies. As an independent benchmarking and performance testing organization, United Software Associates works with some of the world’s leading technology companies and provides valuable guidance to users and developers across industry segments.
The following results reflect a YCSB workload with 50% reads and 50% updates.
Testing Maximum Performance
When configured for maximum performance, each system can potentially lose data when a node fails, which may be acceptable for certain niche applications. This is the default configuration for Couchbase. When all three databases are configured the same way, MongoDB provides 2x the throughput of Cassandra, and 50% greater throughput than Couchbase. All products provide equally low latency in this configuration.
Testing Performance with No Possible Data Loss
When configured for no possible data loss, MongoDB outperforms Cassandra by over 5x and Couchbase by more than 25x, with latency that is more than 95% better than Cassandra, and more than 99.5% better than Couchbase.
Testing High Performance with Minimal Possible Data Loss
In a balanced configuration that provides excellent performance and minimal possible data loss in the event of a node failure, MongoDB provides 3x greater throughput than Cassandra in read-intensive workloads, and 70% higher throughput in write-intensive workloads, while providing 80% lower latency.
This is the default configuration for MongoDB and Cassandra. Couchbase does not provide an equivalent balanced configuration. Instead, users must choose between a configuration that is optimized for throughput, with significant potential for data loss, or a configuration that is optimized for durability, where throughput is 99% worse than the default configuration, and where latency is 200x worse.
About the Benchmark
United Software Associates used the Yahoo! Cloud Serving Benchmark (YCSB) as the basis for its tests, and evaluated three different configurations: one optimized for durability, one optimized for throughput, and a balanced configuration that provides strong performance and strong durability. The latest versions of each product were used: Couchbase 3.0.2, Cassandra 2.12 and MongoDB 3.0.1 with WiredTiger. In all tests the best practices for each product were followed, and multiple rounds of tests were run to determine the optimal number of threads and best performance for each product.
Detailed testing environment specifications and configurations can be found in the full report.