In this podcast episode, Chris reviews the results from a recent project with Ondat, which examines the relative efficiency of container-native storage solutions running common database platforms on Kubernetes. Ondat CTO Alex Chircop and Customer Success Architect Chris Milsted join the conversation.
The work expands on an initial project published in January 2021 that performed synthetic benchmark testing against a range of commercial and open-source container-native storage platforms, including Ondat. This new report provides improved real-world insight, demonstrating the relative performance of MongoDB, PostgreSQL and Redis solutions.
The testing shows how Ondat, Portworx, OpenEBS, Longhorn and Rook/Ceph perform when running database performance testing software. The results measure transactional output including latency and throughput. An extra aspect of this testing is the additional data covering outliers; effectively understanding how well the solutions manage to deliver predictable I/O responses.
In this episode, both Alex and Chris explain the reasons why testing is so important for customers. The relative efficiency of container-native storage solutions varies widely, which can result in additional hardware expense for customers. This is of critical importance both on-premises and in the public cloud.
To access the report, visit https://www.ondat.io/benchmarking (registration required). The original synthetic benchmark testing can be found here
https://www.architecting.it/product/performance-benchmarking-cloud-native-storage-solutions-for-kubernetes-ebook/
Elapsed Time: 00:47:53
Timeline
- 00:00:00 – Intros
- 00:01:00 – Chris is the guest this time! (almost)
- 00:02:15 – The first benchmark looked at synthetic testing (fio)
- 00:03:30 – What were first thoughts and opinions?
- 00:05:20 – What are the performance metrics to be considering?
- 00:07:30 – Device evolution is giving us cheaper and faster storage
- 00:08:15 – Kubernetes storage efficiency should match application density
- 00:10:00 – Performance analysis provides application-specific insights
- 00:11:15 – Operations per second is a better measure of application performance
- 00:12:30 – What happens with 32TB and 100TB drives?
- 00:15:30 – Networking, scalability and failures means there’s a lot of things to manage
- 00:16:45 – Strong consistency and checksums are table stakes for CNS
- 00:19:30 – What did the performance analysis report show?
- 00:21:30 – Creating a fair test environment is a challenging process
- 00:23:15 – This report was able to look at performance outliers and predictability
- 00:26:15 – Ondat spends significant effort in performance tuning their solution
- 00:28:45 – Performance management is an ongoing challenge as infrastructure evolves
- 00:31:00 – CNS will act as a translation layer for applications and storage
- 00:37:00 – What can we learn from the output of this testing report?
- 00:39:45 – What did Chris E find interesting in the report results?
- 00:42:15 – What could we test next?
- 00:47:00 – Wrap Up
Copyright (c) 2016-2023 Unpacked Network. No reproduction or re-use without permission. Podcast episode #edgo.
Podcast: Play in new window | Download
Podcast (storageunpacked): Play in new window | Download