Altoros, a global IT service provider that has been comparing databases for several decades, recently released their latest independent benchmark comparing the DBaaS offerings from Couchbase and MongoDB, leveraging the YCSB standard for NoSQL database benchmarking. This blog summarizes their findings, which prove that Couchbase Capella outperforms MongoDB™ Atlas across various workloads and cluster sizes.
Throughput (how much data transferred from source to destination) and latency (time taken to be transferred from source to destination) were tested on three different cluster configurations of 6, 9, and 18 nodes and executed on the following four workloads:
YCSB Workload A. Update heavily: 50% read and 50% update
- This workload simulates the typical actions of an e-commerce application.
YCSB Workload E. Scan short ranges: 95% scan and 5%
- This workload simulates threaded conversations, where each scan goes through the posts in a given thread (assuming the entries are clustered by ID).
Pagination Workload. Filter with offset and limit
- The workload simulates a selection by field with pagination. Pagination is used for listings, such as e-commerce category pages or search engine results pages.
JOIN Workload. JOIN operations with grouping and aggregation
- The workload simulates a selection of complex child/parent relationships with categorization.
Conducted in June 2022, these tests showed that Couchbase Capella significantly outperformed MongoDB Atlas across all the workloads and cluster sizes measured. The following graph offers a summary of Workload A. Detailed results are available in the Altoros report.
The throughput of each database consistently grew as the number of nodes increased with Capella clearly outperforming Atlas on each cluster configuration. As shown on the graphs above, Capella’s throughput was approximately 10x higher than Atlas for each node configuration, culminating with a throughput of 523,020 ops/sec on an 18-node cluster and latency of 0.8 milliseconds for Capella and 7.8 for Atlas.
Capella features industry-leading performance with a built-in object-level cache, SQL++ query language, ACID transactions, and the ability to scale resources such as CPU and RAM depending on the workload. The Capella query engine supports aggregation, filtering, and JOIN operations without needing to model data for each query. In contrast, Atlas does not support JOIN operations on sharded collections out of the box.
“As in previous benchmarks, Capella demonstrated better performance than Atlas due to its active-active, all-worker nodes architecture,” said Ivan Shyrma, data engineer at Altoros. “Capella is also easier to query due to its SQL support. These factors translate to a better price-to-performance ratio in real-world environments.”
Capella can efficiently process the same workloads with fewer nodes which reduces costs. Based on the benchmark results, Atlas users need to run 18-node clusters to match the performance of 6-node Capella clusters. The monthly cost of a 6-node Capella cluster is $5,284 while the cost for an 18-node Atlas cluster is $28,050. That works out to a savings of ~ 81%. A higher ROI is the practical benefit of doing more with Couchbase Capella on 6 nodes than MongoDB Atlas on 18 nodes.
Learn more about Couchbase Capella:
- Sign up for a free 30-day trial if you haven’t already.
- Connect your trial cluster to the Couchbase Playground or connect a project to test it out for yourself.
- If you’re already using Couchbase Capella, you can interact with your cluster by using the interactive Couchbase shell or via the Capella Control plane to:
- Check out the document viewer
- Connect to a project
- Take a look at the Capella Learning Path!
See the documentation to learn more about the Couchbase SDKs.
Thank you for reading this article. If you have any questions or comments, please connect with us on the Couchbase Forums.