Pure suggests dishing out intelligence to dumb storage shelves
Toshiba thinking 200 layers of 3D NAND
Update At Pure's Accelerate conference last week, the company talked about distributing intelligence to its NVMe fabric-accessed storage shelves.
Pure's all-flash arrays have a dual-controller design, with the controllers talking to dumb storage shelves. The idea seems to be that some controller intelligence could be implemented at shelf level, with the NVMe fabric linking upper-level controller intelligence with lower-level shelf intelligence.
The shelves could then carry out operations on their drives, relieving the upper-level controllers of that work and allowing them to do more, such as cope with a more scaled-up array.
It might also mean that the two controllers could run more upper-level services, such as file systems. It might also lead to having more than two controllers.
There is a whisper that all-flash array vendor Kaminario, which also has an NVMe shelf technology in its road, could be thinking of distributing intelligence inside its arrays in a similar fashion.
Toshiba and flash bits
Toshiba presented on its 3D NAND roadmap at Pure Accelerate. According to Stifel analyst and MD Aaron Rakers, it talked about increasing the layer count to around 200.
Currently Toshiba and foundry partner WDC are at 64 with their BiCS technology. That produces a 512Gb chip. Tripling the layers to 192 would mean a 1.536Tb die. If the two shrank the cell size as well then there could be even more capacity. ®
Toshiba's XG5 NVMe M.2 drive uses this chip and has 256GB, 512GB and 1TB capacities in its NVMe M.2 format. WDC's Blue SSD uses it as well and goes up to 2TB in capacity.
Tripling the chip's capacity from 512Gb to 1.5Tb could mean XG5 technology goes up to 3TB and WDC's Blue expands to 6TB.
Toshiba also discussed its quad-level cell (QLC – 4 bits/cell) technology and said QLC 3D NAND was coming. So let's play with numbers some more. The XG5 and WDC Blue drives are TLC ones (triple-level cell – 3 bits/cell). Implementing them as QLC drives would up their capacity by a third.
A QLC, 192-layer XG5 would have 4TB capacity and a similar WDC Blue would have 8TB. These QLC drives would have quite low write endurance and be only suitable for so-called active access archival applications.
Getting back to Pure, suppose we thought about the idea of a Pure array having NVMe shelves frilled with QLC flash. That would give us an active archive FlashBlade maybe. ®