Keatah

Let me rephrase that, they can ONLY WRITE SEQUENTIALLY. Random writes have to sit in a cache while the HDD clears several adjacent tracks to make room for your incoming data. There is no direct overwriting like in PRML.

That's why they have these big caches. A reasonably accurate way of testing for SMR is to use Crystal Disk info and look to see if it supports TRIM. And the data sheet to see the cache size. 128MB/256MB is a dead giveaway.

For sure. First thing I did about learning about these things was look up the details.

I'm just going to avoid them for now. It seems they're more sensitive to vibration, and I don't trust that they'll have the same lifespan as more traditional drives. Maybe after some more years of other people testing them out for me, I'll take another look.

I got around to reading some technical papers on SMR and the algorithms used to process the translation tables. They appear to operate similar to SSDs - a tiny series of tracks = a block. But with the exception that you can sometimes erase parts of the block unlike an SSD where you have to erase the whole block. You can also sometimes re-write to parts of a block that have already been written to - which is a no-no on SSD's. Similar but different in the precise rules and execution.

Anyhow. Newer SMR drives have like a 50 or 100 GB cache on the platters that is used for immediate writes. And as the drive idles, it does GC and TRIMs itself and organizes and compacts the tracks, thus making free space contiguous. And reducing fragmentation.

I would like to say elegant, but there's a lot of dilly-dallying and farting around going on inside these drives. They are not like the dumb drives of yesteryear where the OS was aware of every file's location.

I think that as time goes on we'll be seeing more and more storage devices with translation layers which make your data's precise location more and more abstracted away from the OS. No longer is there a simple file allocation table. And I understand that Samsung SSDs are NTFS aware and work with the OS in some manner.

I think that's a good thing and a necessary thing as more advanced technologies like HAMR and MAMR with all the lasers and spin torque oscillators. Or like the new QLC flash chips that work with like 8 or 16 voltage levels - that's like analog! Can't expect the OS to be aware of all the gritty details of these black boxes, can we? So storage devices will become more and more complex. Not that today's stuff isn't frightening already.