Michael Dexter
Stefano Marinelli
Michael Dexter and 1 other boosted

A new BSDCan video has been posted:
ZFS Direct IO Benchmarking Pitfalls by Mateusz Piotrowski

https://youtu.be/tYCN8Yg-0JQ

Not too long ago, support for direct IO landed in OpenZFS after years of discussions and reviews. We truly live in the future where we can finally reject complicated caching and fully embrace the unbuffered conversations with our disks. Or can we really?

Those of you who know a bit about ZFS know that the ARC is actually pretty important (without one ZFS would historically stand for zzz 馃槾 instead of Zetta). How could it be then that skipping the ARC might improve performance?

During the presentation we will discuss what workloads and setups benefit from direct IO, what its limitations are, and what pitfalls to avoid during benchmarking. We will also look at the implementation to understand how all the promises of stability and compatibility were kept.

Direct IO is reported to deliver amazing performance boosts in some deployments. Understanding how not to hold it wrong is a great first step to potentially unlocking that speed-up on your systems too!

For more information, please visit:
https://www.bsdcan.org/2025/
- and -
https://www.bsdcan.org/2025/timetable/timetable-ZFS-Direct-IO.html

#zfs #benchmarking #freebsd

A new BSDCan video has been posted:
ZFS Direct IO Benchmarking Pitfalls by Mateusz Piotrowski

https://youtu.be/tYCN8Yg-0JQ

Not too long ago, support for direct IO landed in OpenZFS after years of discussions and reviews. We truly live in the future where we can finally reject complicated caching and fully embrace the unbuffered conversations with our disks. Or can we really?

Those of you who know a bit about ZFS know that the ARC is actually pretty important (without one ZFS would historically stand for zzz 馃槾 instead of Zetta). How could it be then that skipping the ARC might improve performance?

During the presentation we will discuss what workloads and setups benefit from direct IO, what its limitations are, and what pitfalls to avoid during benchmarking. We will also look at the implementation to understand how all the promises of stability and compatibility were kept.

Direct IO is reported to deliver amazing performance boosts in some deployments. Understanding how not to hold it wrong is a great first step to potentially unlocking that speed-up on your systems too!

For more information, please visit:
https://www.bsdcan.org/2025/
- and -
https://www.bsdcan.org/2025/timetable/timetable-ZFS-Direct-IO.html

#zfs #benchmarking #freebsd

Do you want MySQL with the best possible performance? Then stick to version 5.6, as long as you don't mind it's been EOL for 12 years. 5.7 also has better performance than newer versions, although not as good as 5.6. Some performance regressions are just not being fixed.

https://smalldatum.blogspot.com/2025/08/sysbench-for-mysql-56-thru-94-on-small.html

^^ Veteran database performance guru, Mark Callaghan, former tech lead for MySQL at Facebook and Google, is now an independent consultant who blogs about database benchmarking.

(For my own account as a MariaDB fanboy I must add that another option could be migrating to MariaDB which is a plug-in replacement for MySQL 5.6, as MariaDB has done a better job at avoiding and fixing performance regressions.)

#mysql #benchmarking

Do you want MySQL with the best possible performance? Then stick to version 5.6, as long as you don't mind it's been EOL for 12 years. 5.7 also has better performance than newer versions, although not as good as 5.6. Some performance regressions are just not being fixed.

https://smalldatum.blogspot.com/2025/08/sysbench-for-mysql-56-thru-94-on-small.html

^^ Veteran database performance guru, Mark Callaghan, former tech lead for MySQL at Facebook and Google, is now an independent consultant who blogs about database benchmarking.

(For my own account as a MariaDB fanboy I must add that another option could be migrating to MariaDB which is a plug-in replacement for MySQL 5.6, as MariaDB has done a better job at avoiding and fixing performance regressions.)

#mysql #benchmarking