Microsoft Project Denali a Game Changer for Flash Storage at Scale

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

zir_blazer

Active Member
Dec 5, 2016
357
128
43
Isn't this the same thing that LightNVM? It was supposed to be used for cards loaded of NAND chips and a rather dumb controller with a PCIe interface, and what the NAND Controller would typically do was done by the host.
As always, either no one produces the interesing novelty products, or they are nowhere to be found.
 

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
Just like OS controller SMR type spinning disks in a way. Won’t be any use to use more than likely when decommissioned as we don’t have the smarts in the OS.
Probably will work well for them simply due to the scale of Azure.
 

ZeDestructor

New Member
Oct 3, 2015
22
5
3
32
Isn't this the same thing that LightNVM? It was supposed to be used for cards loaded of NAND chips and a rather dumb controller with a PCIe interface, and what the NAND Controller would typically do was done by the host.
As always, either no one produces the interesing novelty products, or they are nowhere to be found.
Looks like it, based on the authors of the most recent LightNVM paper/presentation working at CNEX, who are MS' major partner for Denali. Maan.. I remember musing about the possibility of this back in 2013-2014 when the Intel S3700 was shiny and new and people told me it would never happen XD

Just like OS controller SMR type spinning disks in a way. Won’t be any use to use more than likely when decommissioned as we don’t have the smarts in the OS.
Probably will work well for them simply due to the scale of Azure.
I disagree. With this tech, we can eliminate all the dirty hax they use in SSDs in the name of performance, and instead tune it how we want it to perform purely by changing stuff in the OS/FPGA or accelerator ASIC instead.

Want uber-fast IOPS at the cost of insane amounts of wasted capacity? This can do it (with some help from RAM/Optane caching). Want to use every single last bit of raw NAND and handle all the redundancy in software? This can do it too, using the exact same NAND.

In fact, I don't see any reason why we couldn't just get user-configurable NAND too - buy QLC-capable NAND, configure at deployment point to run in 1, 2, 3 or 4 layer mode, based on needs!