protator

DrivePool lets you pool Raid and StorageSpaces volumes ... so it is absolutely possible to create a Raid 10 equivalent. But it only makes sense for a setup with more than 4 disks/ more than one mirror, otherwise it'd be easier to simply go hardware Raid10 via the mainboard controller (every $80 board from the last decade should be able to do that) or create a striped 2-way mirrored StorageSpace. Onboad controllers usually have enough bandwidth for a 4-disk array and StorageSpaces is pretty robust and performant (as long as it isn't a parity or multi-tier setup^^) so there's no need to go through multiple software layers for a minimalistic array. That being said, it could still be worth considering if you plan to add drives to the pool/array in the near future. Neither a raid controller nor StorageSpaces will to let you change an existing configuration without scrapping the old array and its data first. But if you create Raid0 arrays or "simple" StorageSpaces and add those to a pool in DP, you can easily add more striped volumes later for more capacity or more duplication. You could even upgrade the striped volumes from 2 to 3 or more drives later since it wouldn't affect the mirrors/ duplicated files ... just upgrade one stripe set at a time and let DP add the missing file duplicates back to it. There will be more overhead tho and striping data via software raid does require additional compute power and available threads on the cpu. I am not aware of any concrete minimum requirements, but with StorageSpaces in the mix my recommendation regarding older hardware would be: two physical cpu cores/ four logical cores for the OS, DP and networking plus one additional logical cpu core for each striped ("Simple")StorageSpace in the pool. I know it seems like a lot, and you could probably run all of it on an old core2duo and nothing would explode immediately, but the fewer cores available, the more you'll be relying on Windows resource management to do a good job. Heh, yeah ... you don't wanna do that.

PrimoCache has gained quite a bit in functionality in recent years. L2-SSD caches were rather simple when they were introduced - you could only assign entire disks - but now you can use just some of a SSD's capacity for caching and still create regular ntfs volumes on the rest. I think - never tried it tho* - you can even create several caches on the same SSD. That would give you the option to, instead of using multiple separate SSDs as caches for their respective HDDs, combine the SSDs in a fast Raid0 array and then simply distribute the total capacity between the HDDs you want to cache. (*I did use part of a 4-ssd raid0 scratch disk for caching, so I know that bit works, I just never tried to place a second cache on that array)

O...r, you install PrimoCache, configure a chunk of your SSDs' capacity as an actual block-level cache, and add the remaining capacity to the pool as regular SSD storage.

I just realized ... three M.2 drives ... assuming your system is not based on a HEDT or server platform that would mean it's impossible to run three NVMe drives with four PCIe lanes each at Gen.3 speed. If it's a mainstream platform then it has 16 to 28 lanes tops. At least one of the M.2 connectors is either sharing lanes with the PCIe slots via a PLX chip or, which is more likely, drops down to Gen.2 speed if PCIe slot #1 is using all of its 16 lanes. So check the manual, the Evo Plus might actually be totally fine.

The 'Write-VolumeCache' command in Powershell lets you flush the write cache of a disk, so if you have to unplug it w/o ejecting it, this should prevent garbled data. I haven't tried it myself yet, but apparently there's a "Sync" tool in MS' "SysInternals Suite" that was written to do the same thing

That 970 is certainly not performing as it should. The 970 Evo Plus is one of the best and fastest Gen. 3 SSDs on the market, a 2TB EVO Plus should outperform a 1TB non-Pro 980 or at the very least be just as fast. I've had similar issues with Samsung SSDs in the past, though. All problems with low disk IO were solved quickly by applying firmware updates via "Magician" and (after a reboot) resetting the drive using the "clean" command in Diskpart ('clean' removes all data and partitions from a drive, so don't forget to backup). The latter will most likely cure your limping SSD even without an update since the cause appears to be a misaligned partition in most cases. Not sure how that is even a thing on nand storage, but apparently it is.