A 60GB SLC SSD might have had *gets out fag packet* ~1/4 the NAND cells of a current-generation 1TB QLC SSD (and a comparable inflation-adjusted price, at least back when I bought them), so even ignoring process improvements and giving QLC a flat 1/4 the write endurance of SLC, that's still going to be comparable in terms of TBW if wear-leveling works as intended.The first gen SSDs were typically SLC based SSD. Then can MLC and now there are Quad-MLC or higher. SLC based SSD by their nature are fast and long lasting. These days Quad or higher MLC based SSD do manage to provide lower cost-per-byte but at the cost of speed and life.
It will also have older, slower chips, and likely fewer of them to stripe across for performance. A 14 year old 32nm flash chip is most certainly not "fast and long lasting" in comparison to any kind of modern SSD technology, in fact those drives are not much faster than decent spinning-rust under sequential workloads (what we bought them for was really seek time).
That's not taking into account the firmware bugs and marketing lies (*cough* OCZ) rampant with early SSDs either.
For something more modern, the reasonably priced DRAM-less 1TB NVME drives in my main desktop get hammered with software compiles (Gentoo), host swap with default swappiness, and otherwise get used both as a general work+play desktop and a staging ground for things that need sorting out before going to the NAS (yes, they also have /var on them, which matters not unless you have busy databases there). Zero effort made to reduce writes, nightly TRIM run on all filesystems.
If SMART isn't a lie, and Samsung's warrantied TBW also isn't a lie, the numbers are 33/600TBW in ~3 years... So they should last another fifty years at this rate. They won't of course, but write endurance is still something not worth worrying about.
Lemme see... NVMEs now 3-4 years old: check... IOPS: Near enough exactly what I got then I first benched the system. Throughput: Ditto. Huh.The IO speed that one gets with SSD when it is new is different after say 3-4 years of usage.
Citation needed.
And this is consumer-grade hardware, mind. Server SSDs generally have more endurance and better warranties (and a commensurate price premium).
All this is really beside the point anyway, my phrase was "pathological obsession"... i.e. a fixation on reducing writes to the point it becomes both a waste of time and effort, and a detriment of overall system performance.
Disabling swap (which is near-enough what setting swappiness to 0 does on any recent kernel) is that. It won't save any meaningful amount of NAND lifetime, and will almost certainly hurt system performance.
Ditto avoiding TRIM at all costs. Trimming is nessesary for the controller firmware to do free-space consolidation and wear-leveling, which should actually improve endurance under most workloads, assuming it's implemented intelligently.
Statistics: Posted by steve_v — 2024-11-26 12:43