Re: Performance variation in traditional HDDs?

Sincere thanks for the thoughtful response.

On Mon, Apr 14, 2025 at 4:24 AM Damien Le Moal <[email protected]> wrote:
> Did you mount all drives exactly the same way ? Do the HDD mounts you use have
> proper vibration dampening ?
>
> Generally, HDD performance variations come from different vibration
> characteristics, due to external factors, including noise.
> E.g. a drive in an enclosure that is close to a fan may not perform as well as
> another drive that is mounted further away from the fan.

This definitely could be the/an issue. We have some dampening but it
isn't as good as it could be.

On the other hand, the machines we do the benchmarking on are
identical in physical layout, components, etc, and the performance
characteristics have followed the drives so far. Only one drive is in
a machine at a time.

I think the path forward is to improve the mounting, and then test n
identical HDDs on n identical machines, cycling through the different
machines. This way each drive instance is benchmarked on each machine
instance, and then the results can be compared to determine if the
identical drives really are behaving differently.

> Another thing: what fio command line did you use ? If you did not use direct
> IOs at high queue depth, you can have problems getting the maximum speed.
> Generally, something like this is good:
>
> fio --name=test --filename=/dev/sdX --ioengine=libaio --iodepth=32 \
>         --rw=read --direct=1 --bs=1M

Yes, this is actually exactly the set of options we use for full drive
sequential write (libaio ioengine, iodepth 32, direct=1, bs=1M)