I agree with the sentiment. If I were forced to run USB SSDs, I’d look into a PCI-E expansion with individual USB controllers for 4 USB ports (kind what LTT used ages ago when troubleshooting the Valve Vive or Vive 2 setup).
But by that point, one’s always better-off with a SATA / SAS HBA - unless you have some really specific low-power requirements, which I think USB might fit better (only in terms of power). I’d still avoid USB like the plague (there’s low-power SATA controllers that will be basically negligible compared to what the CPU will use and of course NVME allows the CPU to reach lower power states that SATA, but with a higher price, but also with additional benefits).
Despite that (off-topic rant)...
My experience with the RPi 4 as my main PC running the OS (and data) off of a USB NVME enclosure for 3 years has been stellar. Other than issues when connecting a USB HDD to the pi (locking the system), it’s been rock solid. The only reason I gave up on that setup was desktop needs (lack of arm packages and compile errors on aarch64) and some kinda baseless fears around corrupted data (I was running swap off the nvme as well and I wanted to get some SMART metrics, but I couldn’t ever find a USB enclosure that reports smart to the OS).
But even this anecdote kinda shows that having a usb controller hold up multiple drives is not that great, to put it mildly. I had the usb 2 controller run with a kb + touchpad combo and the usb 3 dedicated to the usb nvme drive, but having that HDD or another USB SSD always resulted in system lock-up. However, plugging in just USB sticks in the usb 3 port and running dd of some isos (obviously read from the nvme) was fine. I can’t say with 100% certainty that it couldn’t have been a power issue - I was using the official pi 4 brick, it might’ve been a bit much to push the NVME and HDD, so again, idk if this was power or usb related.
Nowadays I’d probably set my pi to netboot off of a dedicated ethernet port for rootfs nfs traffic and a usb ethernet for normal data browsing, but TBH IDK if I’d be drawing more power from the constantly on ethernet port and a 2nd eth for internet access compared to a usb cacheless nvme drive (of course with the NAS doing the NFS being negligible).
But I know that diskless computing makes more sense and is more efficient when you have multiple computers running on your network. If you already have more than 3 PCs, it might make more sense to run the OS off of a NAS and potentially have a separate dedicated local drive for data, but for redundancy’s sake, I’d prefer to keep things centralized, where it can be backed up in one go alongside everything else. Obviously there are cases where you might want local disks, e.g. to run a big database, but there’s still advantages to having the OS boot off of the network and use local disks for fast data, since you can dedicate more ports to some crazy fast setup like raid0 or 10 and backup the data at higher intervals, without risking to lose your OS, although even the OS nowadays is replaceable if you can keep track of the OS state configs.
Why am I talking about this on this thread anyway?
Back to the topic, there’s some risks with running a bunch of drives off of single controllers, which might or might not be worth it (definitely not for me), but it’s always better to go with sata or nvme. Now, there’s things one might value more than others, like complete silence without fans and low-power consumption, which might warrant buying nvme powered NAS’es (like the asustor flashstor or friendlyelec cm3588), or might value the money more and instead go usb (it’s kinda common to see usb storage devices being cheaper per gb up to some capacity, e.g. you can get some teamgroup 512GB usb sticks for about $35 - so x8 that it’s 4 TB for $280, which to be fair, it’s about as pricey as a 990 evo plus 4 TB nvme drive, but you can get way cheaper capacity if you go with smaller sizes, like $14 per 256 GB, meaning x16 it’s $224 for 4 TB - but at the expense of real jankiness and potentially data corruption if your USB controllers - plural - can’t keep up with so many drives).
IDK what to say. The only reason I’d go USB SATA SSD or USB NVME (or even the samsung usb t5 or t7 stuff) is if I’d get an ultra-low power device (like my odroid n2+) and it needed local storage (even if unreliable) and if I made backups of that data often. That’s some pretty high bar even then. Sure, it’d use less power than my h3+ (standing at maybe 8W for the whole package IIRC), but it would make sense in certain workloads (particularly battery powered stuff). A NAS running @ 2W would operate for 357 hours (about 2 weeks) off of my bluetti eb3s, vs 8W, where it would only run for 89 hours (~3.7 days) without charging the battery. However, with a 200W solar panel, even when it was really overcast, I’d still get at least 20W in from the sun for about 6 hours a day, meaning that eventually things would kinda balance out even at 8W (and that’s a small battery, 715Wh).
I tried hard to think of scenarios for USB storage, but I still can’t find any that would make it make sense. If I wanted a low-powered NAS and didn’t care about reliability and performance (because I’d be backing up often), I’d go with an odroid m1 or m1s (it has emmc for the OS and 1x m.2 nvme slot for data). If I wanted to do more on the NAS then to server a Samba share, then I’d get a radxa rock 5b (or b+ if I wanted 2x nvme drives) or the cm3588. If I want even more stuff, then straight into odroid h4+ or ultra territory. More than that and we’re getting into “proper” PC builds already. In none of these would USB make sense.
