Thursday, November 8, 2018

What's Happening To Storage?

My only post about storage since May, was October's Betteridge's Law Violation, another critique of IDC's Digital Universe, and their constant pushing of the idea that the demand for storage is insatiable. So its time for an update on what is happening in the real world of storage media, instead of IDC's Universe. Below the fold, some quick takes.

We last heard about roadmaps from my good friend Dr. Pangloss more than two years ago when I wrote:
But curmudgeons like me remember that back in 2013 the Doctor was rubbing his hands over statements like:
Seagate is projecting HAMR drives in 2014 and WD in 2016.
In 2016 we hear that:
Seagate plans to begin shipping HAMR HDDs next year.
Seagate 2008
So in three years HAMR has gone from next year to "next year". Not to mention the graph I keep pointing to from 2008 showing HAMR taking over in 2009 and BPM taking over in 2013. So actually HAMR has taken 8 years to go from next year to next year. And BPM has taken 8 years to go from 5 years out to 5 years out.
Now, to the good Doctor's joy, Chris Mellor reports Seagate is out with a new roadmap:
Seagate's 2018 HAMR roadmap
Seagate has set a course to deliver a 48TB disk drive in 2023 using its HAMR (heat-assisted magnetic recording) technology, doubling areal density every 30 months, meaning 100TB could be possible by 2025/26. ... Seagate will introduce its first HAMR drives in 2020. ... The chart ... shows Seagate started developing its HAMR tech in 2016 and that a 20TB+ drive will be rolled out in 2020.
"Seagate started developing its HAMR tech in 2016" is just wrong. In March 2012 the same Chris Mellor reported:
Seagate has demonstrated heat-assisted magnetic recording technology with 1 trillion bits per square inch
So the real-time slip of HAMR technology continues. It has now taken ten years to go from "next year" to "the year after next".

In NAND so it begins: Micron mounts head-on attack against 10K disks, Chris Mellor reports that:
Micron has started shipping its 2.5-inch 5210 ION flash drive, positioning it as a 10,000rpm disk drive replacement offering much better read access performance for more or less the same price. ... The low cost (low for flash, at least) comes from its use of 64-layer 3D NAND in QLC (4bits/cell) form, and it has 1.92TB, 3.84TB, and 7.68TB capacity points. ... It is heavily optimised for reads over writes, with up to 90,000/4,500 random read/write IOPS; it does random writes at just 5 per cent of the random read speed.
It isn't clear how much the drive will cost in volume. It certainly performs better on read-intensive workloads, but QLC has much worse write endurance even than TLC, so these drives cannot replace 10K 2.5" hard drives if writes are even moderately frequent. And they definitely won't be cheap enough to threaten 3.5" nearline drives for bulk storage.

If the demand for storage were truly insatiable and flash were destined to kill off hard drives, we wouldn't be seeing headlines from Tom's Hardware like this SSD Prices Could Drop Over 50 Percent In 2019 - Report:
According to a DigiTimes report citing "industry sources" this week, NAND flash prices are expected to continue to drop in 2019 after already seeing a 50 percent drop this year. Earlier reports said that SSD prices could fall to as low as $0.08 per GB in 2019.

The DigiTimes report noted that the continued drop in prices seems to be primarily due to SSD manufacturers expanding their production capacity to increase profitability, as well as the adoption of 96-layer NAND technology.
and from Chris Mellor at The Register with Flash price-drop pops Western Digital's wallet: Surprise revenue fall with worse to come:
Revenues in the first quarter of fiscal 2019 ended 28 September, were 3 per cent down year-on-year to $5bn. Profits were $511m, down 24 per cent. Operating cash flow was $705m, compared to $1.133bn last year, a 38 per cent drop.

What fouled up was flash. CEO Stephen Milligan said strength in capacity enterprise, surveillance hard drives and embedded flash products, which each grew revenue more than 30 per cent on the year, "was offset by ongoing declines in flash pricing".
...
[Milligan] added:
This softening demand, in combination with increased flash supply, has led to a market imbalance resulting in a deteriorating near-term flash pricing environment.

We are making an immediate reduction to wafer starts and delaying deployment of capital equipment. These actions will reduce our wafer output beginning in fiscal Q3 2019 [April 2019].
Clearly, the increase in the supply of flash bits resulting from new fabs coming on-line and all the major suppliers transitioning to 96-layer 3D has coincided with a significant reduction in the rate of growth of demand for flash bits. To balance supply and demand, the price of flash bits has fallen, reducing the return on the investment in the new fabs and the 96-layer technology.

This less-than-insatiable demand for bits of storage isn't confined to flash. WD's Milligan reported that:
Client compute disk drives also performed poorly.

HDD revenues in the quarter were $2.49bn while flash revenues were $2.53bn. A year ago the numbers were $2.6bn and $2.57bn respectively, so both disk and flash revenues fell on the yearly compare.

WD shipped 34.1 million disk drives in the quarter, compared to 42.2 million a year ago. Client compute drives dropped to 16.3 million from 20.9 million. Non-compute (retail and consumer electronics) units fell to 11.2 million from 15.2 million last year, while data centre units were 6.1 million last year and rose to 6.6 million – a bright spot.
Many of the markets that consume flash are suffering decreasing unit shipments:
The problem isn't just declining unit shipments, it is also slower increase in GB/unit. As I've pointed out, for example in Betteridge's Law Violation, data isn't uniformly valuable; some data is more worth storing than other data and it gets priority for storage. This implies decreasing returns to increasing storage per unit. Who really needs half a Terabyte in their iPhone Xs? Is the extra 448GB over the base model really worth $350? (That's $0.78/GB compared with DigiTimes' prediction of $0.08/GB next year). Whether the device is a smartphone, a tablet, or a PC the market has saturated both in device and GB/device terms.

The one market with increasing demand is "the cloud". It needs some flash for speed, but the bulk of its demand for GB is for hard disk. Here are the hard disk numbers:
  • WD shipped 19% fewer drives in Q3 2018 as compared to Q3 2017, but 8% more data center drives.
    CategoryQ3 2017Q3 2018Change
    Data center6.16.6+8%
    Client compute20.916.3-22%
    Non-compute15.211.2-26%
  • Seagate has stopped reporting their declining unit shipment numbers. They now only report Exabyte shipment numbers:

    CategorySub-categoryQ1 fy'18Q1 fy'19Change
    Enterprise Mission-critical 2.1 3.0 +43%
    Nearline 25.1 42.5 +69%
    Edge non-Compute Consumer Electronics 13.5 23.4 +73%
    Consumer 11.1 11.2 +1%
    Edge Compute Desktop + Notebook 18.6 18.7 +0.5%
    Note the dominance of nearline drives in Exabyte terms. Given the increase in drive capacity, the consumer, desktop & notebook shipments must have decrease dramatically.
  • Toshiba shipped 23.4M units:
    Category Q3 2017 Q3 2018 Change
    Nearline/hi-cap 1,205,000 1,299,000 +8%
    2.5" enterprise 1,290,000 1,595,000 +24%
    2.5" mobile/CE16,600,00015,720,000-5%
    3.5" desktop/CE5,605,0004,820,000-14%
    Source
    The graph shows that, over time, nearline drives have been comprising the whole of Toshiba’s growth in the enterprise market.

7 comments:

ksec said...

I have doubt how HAMR would work since its introduction in 2008, until someday I see in the real Action. A laser beam to heat up in micro seconds and change bits? I think WD MAMR is a much more practical option, and will be with us in 2019.

Also worth mentioning despite NAND price dropping nearly 50%, it is only back to 2016 level, so if we see another 50% drop by the end of 2019, what it means is that NAND is fallen back to its original trend / projected line. 2016 - 2018 was only a high flux from a longer term of views.

Another thing worth mentioning is the NAND capacity online in China. Lots of Chinese Smartphone were forced to use a percentage of NAND from Chinese Fabs. Even Apple's iPhone uses NAND from Chinese Fabs as well, this will continue to accelerate in 2019 so may be All the Chinese NAND fabs could reach ~7-10% of total world capacity. This is huge in increase in a unit downfall market, further reducing NAND price.

I wish you could comment more on DRAM though.

David. said...

Thanks, ksec.

I believe HAMR works - IIRC in 2010 I saw a Seagate video of a 2.5" HAMR drive working. I think the delays are a manufacturing yield rather than a technology problem.

I expect that MAMR will work too, but if history is any guide it will turn up significantly later than the roadmap suggests.

I'm interested in storage media because I'm interested in long-term data storage, especially the economics. DRAM isn't an issue for that.

David. said...

Although the flash industry has standardized on 96-layer 3D NAND, and are road-mapping 128 layers, Chris Mellor reports that newcomer:

"China's Yangtze Memory Technology (YMTC) aims to leapfrog 96-layer NAND and move straight to the production of 128-layer flash in 2020 ... The company, founded in 2016, has begun bulk-shipping 32-layer NAND and sampling 64-layer 3D NAND."

Note that adding layers is subject to decreasing returns, so the move from 96 to 128 layers with the same cell size delivers somewhat less than a 33% increase in bytes per wafer.

David. said...

HAMR Milestone: Seagate Achieves 16TB Capacity on Internal HAMR Test Units:

"Achieving another major milestone toward volume shipments of its pioneering HAMR (heat-assisted magnetic recording) technology, Seagate announced it has used its advanced HAMR technology to build and test the world’s first formatted and fully functioning 16TB enterprise hard drive platform in a standard 3.5-inch form factor — the highest capacity hard drives ever produced. Seagate is on track to grow beyond 20TB per drive in calendar 2020."

So, they have prototype drives. The key is manufacturing yield, not just working technology. After this long a delay, they have to keep hyping because they still aren't shyping.

David. said...

Despite predictions of exponential demand growth, the downturn in storage shipments continues. Chris Mellor's Weak flash demand and disk sales leave Western Digital scrabbling to claw back $800m a year reports that:

"flash revenues came in at $2.2bn, down 18 per cent year-over-year and disk was down 23.5 per cent to $2.1b. Disk exabytes shipped in the quarter declined 17 per cent on the year."

David. said...

Chris Mellor reports that SK Hynix sees the same flash weakness:

"For NAND Flash, the bit shipments increased by 10 per cent, but the average selling price fell by 21 per cent. ... SK Hynix will push out its 72-layer 3D NAND and wait to pump out 96-layer products. As always, it's all about controlling costs and waiting out the supply glut until demand rises."

David. said...

Backblaze has released the 2018 version of their invaluable hard drive reliability report:

"The Annualized Failure Rate (AFR) for 2018 for all drive models was just 1.25%, well below the rates from previous years"

As usual, it is a fascinating read.