The equivalent exists for solid-state storage as well. It is useful in some contexts to characterize the capacity and throughput of the various firmware subsystems in the same way.
Much of the motivation when I learned it was based in characterizing the storage to identify defects and weird or anomalous behavior that were not disclosed by the manufacturer. Oddities are relatively common. At the most basic level this was useful for tuning I/O scheduler behavior. Just as useful, it allowed us to surface storage hardware issues to customers before they blamed our software.
Probing the properties of storage hardware is a cool but rarely written about area of systems software with many ramifications for managing performance.
A while back I wrote a simple simulation of a spinning hard disk, with the idea being you could tweak the parameters to exaggerate (eg.) track to track seeking and then work out if your filesystem/database/backup/whatever handled it well. I'm going to bookmark this page so next time I'm bored I can update the simulation to be more physically accurate. https://libguestfs.org/nbdkit-spinning-filter.1.html
this takes me back. i used to work on storage performance and understanding the difference between logical and physical geometry was crucial for aligning partitions correctly. modern drives do a lot of remapping internally so the old approaches don't apply the same way anymore.
one thing the article doesn't really touch on is how trim/discard interacts with the physical layout on ssds. would be interesting to see a followup on that.
The equivalent exists for solid-state storage as well. It is useful in some contexts to characterize the capacity and throughput of the various firmware subsystems in the same way.
Much of the motivation when I learned it was based in characterizing the storage to identify defects and weird or anomalous behavior that were not disclosed by the manufacturer. Oddities are relatively common. At the most basic level this was useful for tuning I/O scheduler behavior. Just as useful, it allowed us to surface storage hardware issues to customers before they blamed our software.
Probing the properties of storage hardware is a cool but rarely written about area of systems software with many ramifications for managing performance.
This study is so cool.
A while back I wrote a simple simulation of a spinning hard disk, with the idea being you could tweak the parameters to exaggerate (eg.) track to track seeking and then work out if your filesystem/database/backup/whatever handled it well. I'm going to bookmark this page so next time I'm bored I can update the simulation to be more physically accurate. https://libguestfs.org/nbdkit-spinning-filter.1.html
Why content like this doesn’t supersede the AI content, I will never know. Thanks for sharing :)
AI, as we have today, struggles to produce long, data-dense article.
this takes me back. i used to work on storage performance and understanding the difference between logical and physical geometry was crucial for aligning partitions correctly. modern drives do a lot of remapping internally so the old approaches don't apply the same way anymore.
one thing the article doesn't really touch on is how trim/discard interacts with the physical layout on ssds. would be interesting to see a followup on that.