NAND I... will always love you, says Micron as it emits 7.68TB QLC SSD Micron has whipped out the 64-layer 5210 ION SSD – apparently the industry's first QLC NAND flash drive – opening a new front in the SSD-HDD marketing war. QLC (quad-level cell) flash stores four bits per cell, one more than today's TLC chips that use three bits per cell. The capacities of this 2.5-inch product are 1.92TB, 3. …
This will be very helpful for HCI vendors with journaling filesystems because they can level writes for the persistent storage tier. We can also expect storage array vendors to take full advantage of it after they have tuned their write caching/de-staging a bit.
In terms of write-caching, these drives are a near-perfect fit for ZFS use.
Forget the L2arc (it's not needed when seek times are near-instant) and put in a high performance SLOG for ZIL - which vendors have been pushing for spinning use anyway (usually things like 8Gb ZeusRAMs, when the most data you'll ever see in a ZIL is a few hundred MB and even that's exceptional)
(We found the hardest part about ZFS was finding vendors who didn't attempt to cripple it by forcing the disk activity through some kind of RAID controller in the mistaken belief that it would improve performance or that battery backed write caches would make things more reliable. There are a lot of sub-par ZFS installations out there and traditional storage vendors are the worst offenders for ruining performance/robustness)
This is where it gets really interesting.
Despite their MAID capabilities my existing nearline arrays have to keep powered up pretty much constantly due to the time it takes an array to come online when needed _AND_ the risk that a 3-5yo spinning drive subjected to many start-stop cycles simply won't spin up when commanded.
As such whilst the gains can be measured in terms of "time to ready" response, the real power savings are an order of magnitude higher than the simple difference between spinner and sdd consumption. These drives can be left in low-power idle state a lot more of the time than spinners can, which means my 900W per shelf continuous load doesn't just drop to 400W, but when averaged over the day comes out somewhere nearer 40.
less heat == less cooling load and lower overall power requirements (which is a big deal in places like central London, where if you want a few hundred extra kW capacity in XYZ location you may have to wait 20 years before the power company can provide it.)
I'll wait to see pricing, but it's the above kind of thing which justifies buying these devices when they may be 3-5 times more expensive than the equivalent spinning media (experience is also showing that
as long as correctly specced for the jobs at hand they last in-service far longer than the HDDs they're replacing. We have decade-old SSDs in service in some (non-critical) corners of the network which are showing zero signs of degradation whilst you'd be nervous as hell about spinners of that age.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
