NAND, DRAM, SAS/SCSI and SATA/AHCI: Not Dead, Yet - EnterpriseStorageForum.com

NAND, DRAM, SAS/SCSI and SATA/AHCI: Not Dead, Yet

Manufacturers are coming out with new non-volatile memory (NVM) media like 3D XPoint. Does that mean that DRAM and other NVM media such as NAND flash are now dead?

Do new NVM storage access protocols such as NVM Express (NVMe) mean SCSI/SAS and AHCI/SATA are now dead?

My simple answer is no, they all have bright futures.

Server and Storage I/O Media Hierarchy

Storage Media Hierarchy

If you need a refresher on NVM (NAND flash, SSD, 3D XPoint) and NVMe, there is a primer on my blog. While some might argue that NAND flash, SAS/SCSI and SATA/AHCI are dead because there is something new, I say hold off on scheduling funerals for a while.

It’s fashionable for some creative marketers and pundits to declare something existing as dead when something new is born. What’s entertaining is when something that was just born or created is declared dead at the hands of something that does not even exist yet.

In my opinion, NVMe has a very bright future, and I’m a fan of the technology. But the reality is that SCSI-based SAS and AHCI-based SATA will be around for several years—although their roles may change. They will be necessary for compatibility with existing servers, storage, I/O devices, hardware, software, systems and deployments, as well as for near-term low-cost high-capacity commodity interfaces.

For example, while NVMe will initially be used for high-performance, low-latency server storage I/O, SATA and SAS will be used when lower cost, lower performance and higher capacities are needed for web-scale, cloud, HPC, big data, video and similar environments. Eventually, NVMe will pick up more of that activity, similar to how SAS picked up more activity from parallel SCSI and Fibre Channel devices. In other words, they will co-exist as part of tiered storage access options.

Example of SATA, SAS and NVMe coexisting for server storage I/O access

Coexistance

The same thing will happen with the new NVMs, such as 3D XPoint, which was recently announced by Intel and Micron. Particularly with new vertical and 3D implementations, NAND flash will extend the capabilities of flash-based solutions for several more years. This is similar to when the HDD was declared dead (again) in the early to mid-2000s when the manufacturers flipped from horizontal to vertical or perpendicular recording. The result was pushing back the super parametric barrier (e.g. the infamous technology brick wall) to increase densities even further, similar to what’s happening today with NAND flash.

It's true that the higher-capacity NAND flash is not improving on read and particularly on write performance at the same rate of some newer NVM such as 3D XPoint. However, keep in mind that 3D XPoint was just announced about a month ago while NAND flash was announced twenty years ago. NAND flash continues to evolve and still has plenty of life, even as new memories evolve. The same is true with DRAM, for reasons like cost, backwards compatibility, cost, availability, cost and cost.

It has taken NAND flash a few decades to mature to where it is today, but it has plenty of life left. Likewise, while NAND flash and DRAM continue to get leveraged, the newer NVM technologies will continue to advance and will coexist in some scenarios.

In the same way, the newer NVMs will co-exist with various NAND flash and DRAM, both for server memory and storage SSDs, at least for several years (if not more).

Here's the reality: NVM is in your future. Some day HDD, tape and even SSDs as you know them will be gone, just as paper tape, core memory and other things are now seen in museums instead of mainstream use. Likewise, NVMe is in your future. The questions are when, where and with what.

Photo courtesy of Shutterstock.


Comment and Contribute

 


(Maximum characters: 1200). You have characters left.

 

 

Storage Daily
Don't miss an article. Subscribe to our newsletter below.

Thanks for your registration, follow us on our social networks to keep up-to-date