Btw, this is excellent advice. Funny story, I have a FreeNAS device and the motherboard died and I thought "Oh my, I need to bring my ZFS volumes up on another machine, but I didn't have another machine with 6 SATA bays! I ended up having the drives all sitting out on the workbench connected a mainboard with 8 SATA ports so that I could create an archive of the data, and then got the mainboard fixed so I could re-assemble and re-use the FreeNAS but still it alerted me to the fact that I really needed a 6-8 drive cabinet if I wanted to do this again.
Hardware raid is legacy :)
Sure, for personal usage there is almost no usage for the HW RAID, but when you need to make sure what the system would always boot and it can't be serviced in hours/days - then you have almost no options for SW RAID.
No problem to put /boot on a raid1 on a small partition across all drives, so that any drive can boot, and no problem to even include a whole self-contained remotely accessible recovery os. It's a little bit more work to set up, but if you are professing a need for that, then a little extra setup is de rigueur. I remotely administered a ton of linux boxes in racks scattered across the US like that for years. Although I had out of band serial console access and could do full bare metal reinstall that way, I could also do it from any neighboring machine that was still running in the same rack if I had to, with a combination of network booting and/or booting from any one of the normal drives normal raid1 copies of the /boot partition.
Further remote-able fallback options that I never even had to use but could: Local hands just plucks a hot-swap drive from any of my other machines and pops it into the bad machine. All drives had the same bootable partition and all drives are redundant and so they could yank literally any one from the wall of server fronts. Or, better, local hands just plugs in a thumb drive and I take care of the rest. Thumb drive is already sitting there for that purpose, or they could make a new one from a download. But with 8 to 24 hot-swap drives per machine, meaning 8-24 copies of /boot, I never even once needed local hands to so much as plug in a thumb drive.
There is just no problem at all with sw raid. It only provides options, not remove them.
Did you even read my comment? It's quite clear what your environment was in the data-centers, with spares and remote hands.
Mine wasn't and then I say three months I don't kid or jest.
> No problem to put /boot on a raid1 on a small partition across all drives
This is exactly the problem. If the drive isn't totally dead (like it doesn't even respond to IDENT) then there is a chance what the BIOS/UEFI would try to boot from it and even succeed in that (ie would load the MBR/boot app) and then there is no way to alter the boot process at this point. HW RAID card provides a single boot point and handle the failing drives by itself, never exposing those shenanigans to the upper level.
Like sure, you are happy with your setup, you never had a bad experience with it, you always had OOB management and remote hands - but it doesn't means what it a silver bullet working 100% of times for everyone.
Yes, I've seen systems with SW/fake RAID failed to boot because the boot process failed after selecting a half-dead drive as a boot device, with my own eyes. Thankfully I was geographically close to them, not 5000km away.
Yes, I serviced and prepared systems for the 5000km away divisions and they are really serviced only a couple of months in a year, all other time you need an extremely urgent reasons why do you need to a rent a heavy 'copter to go there. No, there is no remote-hands there. The maximum point of IT-competency there is raking bills with satellite Internet.
It’s probably time for me to figure out converting over to software raid.
Thanks for the nudge.
I bought four cheap SATA-USB3 adapters, plugged them into two USB3 hubs, which I then plugged into the two USB3 ports of a Raspberry Pi 4, and arranged it all quite precariously in a small cardboard box that I cut holes into for airflow. Performance was terrible, of course, but it worked well enough until I could build a proper new NAS box.
With generic software raid (mdadm), even if you only had a single usb port and a single internal drive, you could image all the drives one at a time and then access the array of images on the single big drive. (not uncommon since usually time has passed by the time something fails)
It also goes the other way if you needed it to. Say a single array member was 4T but you only have a bunch of 1T drives, no problem, you can assemble 4 1T drives into a 4T container to hold the 4T image, and then use that image as an aray member itself.
Even if you don't have any loose external drives, you could even do it all via network shares with pieces residing on all of your other family members laptops and desktops, or every desk in an office, while they all still continue running windows and doing their normal jobs I mean.
Some of the possibilities are slow or fragile or both, so of course you don't set out to use 12 usb2.0 ports, but the point is essentially anything is possible, and you don't have to worry about predicting or planning for every possibility, you just don't have to care about how you'll recover the array in the future because it doesn't matter what form storage takes at that time, or what form you happen to have available. It would almost never make sense to do some things, but the point is that mdadm just doesn't care.
For a machine with only 2 or 4 internal drives where you want to use raid0 for max throughput, and don't want to rely on any special firmware support for booting raid0, just partition the drives so that /boot is a small raid1 across all the same drives, so that any of the drives could boot. Bonus, it automatically makes all the members of your main raid0 slightly smaller than the drive's nominal size, which means you can always fit them onto some other replacement drive later, even if the different manufacturers count bytes and formatting overhead differently.
I come from the days of sco unix on scsi hardware raid with full featured expensive cards and I do not miss it.