In 2014, in a presentation I’ve done, I’ve said to people that in 2-3 years new and cheaper flash memory would have become the standard solution for general purpose disk storage, thanks to a price per GB comparable with spinning disk. Seems that I was right after all.
The prediction in 2014
I’m not a fan of analysts usually. They tend to confirm obvious things that people in the market are already observing since months, if not years. Don’t get me wrong, theiy do have a market, but their customers are usually high-level managers who are either lazy to follow news, blogs, announcements and trends, or that believe something only when it’s said in an expensive report instead of a blog.
However, there are also very smart analysts that bring real value to the business, by making good predictions. We usually have a short memory, so unless we keep track of all those researches and go back after 1-2 years to verify if they were true, we forget those both the wrong researches,and the correct ones.
I remember the source I’ve used in my presentation in October 2014:
I found this graph in an article published at The Register, titled “Flash could be CHEAPER than SAS DISK? Come off it, NetApp” by Chris Mellor in August 2014. In the article, Chris linked this chart that was originally presented by NetApp at Flash Memory Summit 2014. There, they said “the cost-per-GB price trends for cheap TLC flash mean it will dip below SAS disks in 2017″.
The reality in 2015
If you look at the graph, the crossing was supposed to happen in 2017 with a price per GB of around 9 USD. But NetApp research didn’t take into account the incredible pace of innovation occurring in NAND memories!
In previous years, a new technology has been introduced: 3D NAND, and already in 2013 Samsung (one of the few vendors with this technology available in quantity) announced the start of proper production. What probably took NetApp and many analysts by surprise was the short amount of time it took to Samsung and other producers to drive down costs to the point where now many storage vendors can offer flash memories in their storage with a price of 1.50 USD per GB. Probably also thanks to huge cloud providers like Amazon adopting insane amounts of NAND memories (read Is AWS an all-flash storage vendor?), thus driving costs down even more.
At this price, SATA disks are really in sight in the short future, but most of all SAS disks have been already totally defeated! Have you seen lately SAS disks around? No. And here you can see the reason why: it’s not anymore a discussion around price per IO compared to price per GB; today, even pure price per GB of new NAND memories can be compared to spinning disks.
The future in 2016
For sure, 2016 will be a turning point for NAND technologies. I suspect at this point that during next year, the raw price per GB of NAND will become lower than SATA disks, and at that point there will be no reasons left to adopt spinning disks.
How will the market change?
Spinning disks has become in latest years “the new tape”, thanks to its lower price per GB. Solutions for data retention have leveraged spinning disks to offer a second tier of storage for less important workloads, or for backups. But if now NAND memories are cheaper than even the cheapest disk, can we think about having backup medium built using NAND? Maybe yes, while faster NAND memories (non-3D, SLC and MLC, NVMe, and other technologies that will surely come out) will be the new medium of choice for production workloads. With additional benefits like lower power consumption and less moving parts that can break.
Also, another impacted business is hybrid arrays. They have always promoted their value but stating that only the active workload of an application needs to stay on Flash, while the rest of stored data can be saved on spinning disks to reduce the overall price. But if now NAND memories are cheaper than disks, we will maybe see hybrid machines where the tiers are built using different types of NAND rather than flash and spinning disks. We have already seen the arrival in the market of these solutions (both Dell and Coho Data have presented all-flash arrays where two different tiers of NAND are used), while for “traditional” hybrid arrays like Nimble it’s probably time to move to an all-flash configuration.
Surely, 2016 will be another interesting year for the storage market.