On a low-power CPU, overhead for USB protocol handling may be less efficient than PCIe DMA?

RAID seems very popular for 2 disk setups, but I don't like the cost benefit. With RAID under about 5 disks, you aren't getting much speedup, under 4 disks you can't do a proper fail-out and rebuild of a larger multi-disk volume.

I use 2 x 4TB disks in my homeserver, but I keep one online and have a script that brings up the other one periodically and rsyncs everything. This gives me a local backup, something RAID lacks, so I'm protected from fat fingering or accidentally deleting stuff. I also have very minimal downtime, because I can mount that drive in the place of the primary drive in just afew seconds.

I run xfs on the primary drive, and btrfs on the mirror, so I can take snapshots after I rsync and maintain differentials easily.

My point is, you should consider getting rid of RAID and just use the bare drive or LVS Volume.

Agreed, RAID makes little sense if you want a quiet home storage that is used only occasionaly. Additional disk is better used for regular backups. If you need to minimize downtime and have the storage accessible 24/7, then RAID makes sense.

IIRC, SATA over USB3 can offer competitive performance for sequential I/O, but PCIe will give better random I/O performance and lower CPU usage.

tkaiser of the armbian project has a lot to say about this stuff. I find his posts in the cnx-software.com comments and armbian forums very helpful.

https://forum.armbian.com/topic/8097-nanopi-m4-performance-a...

https://github.com/ThomasKaiser/Knowledge

PCie 3 x1 is 8 gigabit/s. But yes you are right, the gigabit network should be the bottleneck either way.

For transfers, yes, but for housekeeping activities such as scrubbing, defragmenting, indexing, and deduplicating, not at all.