Benchmarking Resources for Kubernetes

TL;DR: Always test your own application in an environment representative of the final target - this is always the best scenario to achieve comparable results to what is expected in a production environment.

While there has been a revolution at the hardware layer, including the emergence of NVMe and NVMe-oF, and in the way that workloads are loosely coupled and scheduled with the help of Kubernetes, benchmarking performance still comes down to being as close as possible to the real world experience of your users and their workloads.

One approach that we take at MayaData is to explicitly compare the performance of our software - and the NVMe-oF capable OpenEBS Mayastor in particular - to the theoretical maximum performance of underlying devices and cloud volumes. You can see this approach in benchmarking of OpenEBS Mayastor with our friends at Intel. It is our hope that this approach will provide somewhat of a common denominator for performance comparisons. We also find that users appreciate this - they can trade off the benefits of our software vs. the risks inherent in simply using the raw disk and local persistent volumes without the management or resilience advantages offered by OpenEBS Mayastor.

A second approach we take is to offer tools that assist in creating this real world environment. In particular, we authored an automation playground that is extensible. It can deploy Kubernetes itself and relevant workloads and relevant benchmarks for those workloads. Please take a look and contribute your thoughts and help this open resource grow in value.



The basics and more are available from Storage Network Industry Association.


Kubernetes Storage Performance Myths.


Draft of whitepaper from the CNCF Storage SIG.

benchmarks of OpenEBS

Recent benchmarks of OpenEBS include the following:



2ndQuadrant PostgreSQL (now part of EDB)



Common workload benchmarks:





Have Questions

Have a question on benchmarking or want to talk to us to test your application?