joesiv

One thing that bothers me about SSD's is their specs for lifetime writes, or drive writes per day. rated for 200TB's? Is it really? What does that mean? You would think that doing the math is simple, you take what you typically write to disk, and extrapolate how many years it would take to write that much data? You'd be wrong! Fact is, what you "write", or commit to disk, is never 1:1 what the SSD, or controller/firmware is writing to NAND. And that's the key, the total drive writes per day on the specs is based on NAND writes. Problem is, depending on the controller, and often more importantly the firmware, the results can vary wildly. One of the most simple examles is a user writing a small 1K update to a file. 1K, easy... NOPE, problem is NAND writes are done on a page by page basis... so even for the smallest write, you still have to write a whole page. How big is a page? usually 512KB. So the user writes 1KB, but the NAND writes 512.... That's just one example. Firmware also has garbage collection, and other features which optimize your wear, not only as you write, but to clean up, which leads to more writes, without you actually doing any writes. This is often called Write Amplification Factor (WAF), it's not uncommon to have 8x's WAF, of course this is highly dependent on workload. I was tasked to evaluate several SSD's from various manufactures, and they all had similar specs, however, the firmware was what killed it for most of the drives. Some vendors seem to have very buggy firmwares, and typically go through many iterations before they get it right... on shipping products... Another vendor a tier one memory company had great specs, but for the life of us we could get the life expectancy that they were claiming, something weird in their firmware. This is something that I find noone talks about, and they just assume, wow I can get 2 whole drive writes per day! I wish linus would have touched on that... If you want to really get a sense of how your SSD is doing, and maybe it's doing fine, look at your SMART data, look up the documentation for your SSD, and look for "Average Block Erase Count", it will be a different attribute number depending on the vendor/drive, but most have it. Some you have to decode, which is pretty simple with windows calculator. Then take the rated Cycle count for the NAND, for MLC it's often like 3000 writes (sometimes listed in the specs, but you might need to dig, or talk to the vendor, another thing I wish the reviewers would touch on). Wait a week or something, take another reading. Then start doing some math. If your average block erase count has went up by X amount in 1 week, calculate out how long it will take for that count to get to 3000, easy peasy.... it will likely be a lot less time than you first estimated. (edit) You may be tempted to use some SMART attributes like Total Writes, but usually that is what is commited to the controller, not what is actually sent to NAND, some vendors will have something like FTL_writes, which is NAND writes, but there is no standard, and sometimes aren't exposed at all, and you have to talk to your vendor to really find out what the various SMART attributes mean, as often they aren't documented in the SMART literature for the drive) \rant Oh, on the actual topic, I would be scared even using TLC SSD's, man SSD's are complicated animals, they seem simple, just write to a chip, simpler than spinning drives right? It seems the SSD market really is the wild west, whether it be pushing for greater capacities for cheaper (at the expense of endurance), big storage space numbers sell I guess, or the firmwares that actually do the heavy lifting, and which can make or break (literally) a drive in terms of life expectancy and performance (are we trading off performance for endurance here too?) Performance sells afterall... I think things are getting much more mature on the firmware side of things, but the hardware, we are really getting nowhere, since you are sacrificing life for space... it makes me sad.