For optimal value from Optane-like devices (byte-addressable, high performance) which you want to use as persistent storage (as opposed to, eg, using it as a paging store or a RAM disk) you need a file system (or equivalent like a database).
A traditional file system *will* work (ie emulate flash) but is sub-optimal for two reasons:
- it goes through the OS and layers of abstraction. For traditional storage the latency of these layers is negligible, but it's a significant cost when the distance to storage is a small multiple of the distance to DRAM.
- traditional file systems are designed to work with large chunks (eg 4kB size) and so do not take advantage of byte (or more realistically cache-line) addressing.
Work is being done on both these fronts. What is known publicly is Microsoft is working on a file system that is optimized for these characteristics like byte-addressable, and work is being done in Linux on how to make the control layer between app and storage as thin as possible.
A realistic analysis today, I think, would be that on the software side there are no lock-ins. I would expect that everyone you'd think has an interest (from Apple to Oracle to IBM) is working on one or both of the aspects I listed above; they just haven't said anything.
On the hardware side, hmm. Here's my analysis.
Intel, now in its decadent phase, is being run increasingly by a finance mindset (look for rent opportunities and exploit them as long as you can) rather than an engineering mindset (spread new tech as widely as possible because somehow it will take off and smart people can make money from that). So everything around Optane was designed to force enterprise customers to buy expensive special Xeons and expensive special Optane DIMMs, with no interest in the wider world beyond deep-pocketed enterprise.
Micron, still an engineering company, considered this an idiotic, short-sighted strategy, and so took back as much control as possible over the tech, and wants to sell it like any new tech you want to take off, at a range of price points, with huge ambitions that it spread everywhere (just like flash).
So where does that leave us? ARM v8.2 (2016) included the sort of cache control instructions needed to control persistent memory, and presumably these will trickle out over the next few years into ARM server and lower end cores. (Apple definitely uses some ARMv8.3 features, but I don't know for definite if anyone has implemented these cache control instructions yet.) Presumably at some point AMD and IBM will do likewise, but I have seen nothing from them yet.
So, for another year or three, likely Intel will have the game to itself but fairly soon (before 2025, maybe 2023) I expect a commodity spec for the DIMMs, Micron to supply them (with, who knows, maybe Nantero and MRAM and other weird options also providing DIMMs), all the relevant CPUs providing the instructions necessary to exploit this stuff, and all the relevant OSs providing their part.