Comment by w10-1
1 day ago
Optane persistent memory had a fascinating value proposition: stop converting data structures for database storage and just persist the data directly. No more booting or application launch or data load: just pick up where you left off. Died because it was too expensive, but probably long after it should have.
VM's persist memory snapshots (as do Apple's containers, for macOS at least), so there's still room for something like that workflow.
1+ for 3dxpoint.
The technology took decades to mature, but the business people didn’t have the patience to let the world catch up to this revolutionary technology.
The world had already caught up. By the time it was released, flash memory was already nearing it's speed and latency, to the point that the difference want with the cost.
>flash memory was already nearing it's speed and latency
Kinda, but for small writes it's still nowhere near.
Samsung 990 Pro - IOPS 4KQD1 113 MBytes/Sec
P4800X optane - IOPS 4KQD1 206 MBytes/Sec
And that's a device 5 years newer and on a faster pcie generation.
It disappeared because the market that values above attribute is too small and its hard to market because at first glance they look about the same on a lot of metrics as you say
Optane was impressive from tech standpoint.
We were about get rid of split between RAM and disk memory and use single stick for both!
Systems are stuck in old ways in how they model storage, so they weren't ready for something that is neither really RAM nor disk. Optane did inspire quite a few research projects for a while though. A few applications emerged in the server space, in particular.
I have an optane drive with the kernel on it, instant boot!
How does that work? It loads kernel from drive to ram?
Isn't windows fast boot something like that (only slower, depending on ssd)? It semi-hibernates, stores kernel part of memory on disk for faster startup.
This one would have behaved more like suspend to RAM. In suspend to RAM, the RAM is kept powered, while everything else is shut down. The recovery would be near instant, since all the execution contexts are preserved on the RAM.
Optane was nearly as fast as RAM, but also persistent like a storage device. So you do a suspend to RAM, without the requirement to keep it powered like a RAM.
Not only because of price. The 'ecosystem' infrastructure wasn't there, or at least not spread wide enough. The 'mindshare'/thinking of ways how to do, neither. This is more aligned with (live) 'image-based' working environments like early Lisp and Smalltalk systems. Look at where they are now...
A few more thoughts about that, since I happen to have some of the last systems who actually had systems level support for that in their firmware, and early low-capacity optanes designed for that sort of use. It's fascinating to play with these, but they are low capacity, and bound to obsolete operating systems.
Given enough RAM, you can emulate that with working suspend and resume to/and from RAM.
Another avenue are the ever faster and larger SSDs, in practice, with some models it makes almost no difference anymore, since random access times are so fast, and transfer speeds insane. Maybe total and/or daily TBW remains a concern.
Both of these can be combined.