Linux 5.1 Might Pick Up Support For Using Persistent Memory As System RAM

[phoronix]

Phoronix: Linux 5.1 Might Pick Up Support For Using Persistent Memory As System RAM

While we are expecting to see more Intel Optane NVDIMMs this year that offer up persistent memory using 3DXPoint memory on the DDR4 bus for persistent storage, the Linux 5.1 kernel might pick-up support for treating this persistent memory back as traditional RAM if so desired...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-5.1-DAX-PMEM-AS-RAM

[sireangelus]

can someone explain to me the difference between this and swap?

[HadrienG]

Lower latency because there are less OS layers to go through, I believe.

[pegasus]

PMEM or persistent memory is a storage device that happens to be byte-addressable (load/store). Traditional rotational media are block addressable (read/write) and paging space (=swap) is designed to not retain data across reboots. Pmem can retain it, it's actually a bios setting to decide what to do with data on reboot. It will take some years for OS and apps to get used to this new reality (ram that survives reboot) and start making use of it. Browse pmem.io for programming samples.

[Slartifartblast]

It will take some years for OS and apps to get used to this new reality (ram that survives reboot) and start making use of it..

Hmmm, persistent memory that survives reboots, that's a tasty malware target. Let's hope Intel make a better job of security than the cluster fsck they did with their CPU's and IME.

[schmidtbag]

Hmmm, persistent memory that survives reboots, that's a tasty malware target. Let's hope Intel make a better job of security than the cluster fsck they did with their CPU's and IME.

Good point. At least in this case it's pretty much all in software. So at least for Linux and other open-source OSes, many eyes can look at the source code to determine security flaws. Of course, this also means the attackers get to spot problems early too without reporting them. Windows users are the ones who really need to watch out, especially considering how sloppy MS has been with their updates in the past year.

[Weasel]

I don't know much about Optane but aren't they technically SSDs, so have a limited amount of write cycles? RAM gets written to a lot compared to anything on disk, and usually in very small amounts as well.

Yeah I'm probably wrong, but I wouldn't be comfortable even with tens of thousands of cycles for this, especially since once RAM gets corrupted or "fails" to write to, shit hits the fan.

[pegasus]

Optane endurance is a hot topic these days, especially since Intel didn't say anything officially yet. I have some sources where I can get first-hand info, but lets wait until official unveil which will be together with cascade lake xeons I presume.

[AndyChow]

I don't know much about Optane but aren't they technically SSDs, so have a limited amount of write cycles? RAM gets written to a lot compared to anything on disk, and usually in very small amounts as well.

Yeah I'm probably wrong, but I wouldn't be comfortable even with tens of thousands of cycles for this, especially since once RAM gets corrupted or "fails" to write to, shit hits the fan.

One early optane SSD had a rating of 32k erase cycles. Intel is currently pushing for a 60 DWPD rating. But that's a lot less than initial projections, which is why we are where we are. But yeah, it's hush-hush, and optane is not performing to expectations. Remember, it was supposed to be 1000x faster than NAND, and have 1000x the endurance of NAND, and currently it's not even close to either of those numbers.

[pal666]

can someone explain to me the difference between this and swap?

you can't work with swapped out memory, you have to swap it back (swapping out some other memory to free space). i.e. all you are doing is useless ping-pong. subj just reads/writes cachelines as usual.