A little bit of TLC: How IBM squeezes 16,000 write-erase cycles from QLC flash IBM says it has managed to coax TLC-class endurance and performance from cheaper QLC flash chips, with customers of the company's FlashSystem 9200 all-flash arrays getting the benefits. No one else can do this, according to Andy Walls, IBM Fellow and CTO for flash storage products. Quad-level cell flash is cheaper to make …
You won't get even 1.5x with, say, video...
On-device compression is largely a marketing con. One also has to consider the interface bandwidth issues: a NVMe/PCIe3 x4 has a theoretical throughput of approximately 4GB/s. There are multiple vendors of SSDs that can sustain speeds (at least for bursts lasting 10s of seconds) in excess of 2GB/s. So even if you did could achieve 3x compression on the data, you won't realize it. Better to put the compressor on a controller...
The very last line in the article, about the cost of characterizing the devices, seems wrong. I thought the point of the invention was that the controller does the characterization autonomously, early on in the device's service life. So it's not taking up time during manufacturing, and thus not incurring cost in the usual sense. Did I misunderstand?
Having flash that lasts through 16 times as many writes as it might is good. But caching writes to TLC or QLC flash and only writing
out cached contents to flash infrequently also is a way to extend the life of flash storage. For caching with DRAM, if one has battery backup for
when power goes out, then the caching works o.k. Like the battery of a laptop, or battery backup for servers.
But if there is no battery backup, having a lot of flash disk updates in DRAM cache when power goes out is problematic.
Consumer flash drives don't say much about caching when one buys them.
I think the manufacturers are working their spec sheets very carefully instead.
Seagate "Barracuda" flash drives for consumer upgrades are rated at total writes of 600x the capacity of the drive. Which is pretty good as long as you don't do video editing all day long, but spend your time reading El Reg instead -- meaning the disk is idle.
Enterprise drives are about 3x better write endurance, if my math is approximately correct (Nytro 1351 series, just to pick a product.)
The end user can also adopt these strategies: keep the drive less than half full; use big RAM as a buffer to reduce writing rate; sleep more often.
Some filesystems let you disable all synchronous writes and buffer flushing (as long as you have RAM to spare). It really cuts filesystem latency and storage writes on cloudy instances that don't exist after a reboot anyways.
"Quad-level cell flash is cheaper to make than triple-level cell flash and increases storage density"
Is QLC actually cheaper to make than TLC? My impression was that it's cheaper per bit stored, but it's almost certainly more expensive to make per wafer, simply because it's more complicated. So it's not that it's cheaper to make and increases storage density, but rather that it's more expensive to make but increases storage density by more than enough to compensate for that in a finished product.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
