EMC: flash replaces high-end disks in 2010

by Robin Harris on Monday, 19 May, 2008

Greetings from Las Vegas
And EMC World 2008.

Dave Donatelli, president of EMC’s storage business, presented to the press room this morning. His most interesting statement was that flash drives will have cost-parity with, and therefore replace, high-end rotating magnetic disks, by the end of 2010.

Let’s run some numbers
Dave said that EMC has measured STEC’s flash drives at 30x the IOPS of a high-end disk with sub-millisecond access times. That alone would justify a premium over existing drives. He also said that the performance of the flash drive was better under load. A double win.

A 15k 74 GB Seagate SAS drive is about $175 or roughly $2/GB. A 2 GB Single Level Cell (SLC) flash chip is currently about $8/GB on the flash spot market. If flash keeps dropping at 50% a year they’ll be where the current disk price is in mid-2010.

But that’s raw chip vs finished disk
The remaining question is how much does the chip controller and other infrastructure cost? STEC isn’t selling its 74 GB flash drives for $8/GB – $80/GB is closer to the mark. Volume should amortize their engineering costs. PC boards are cheap.

That leaves the flash translation layer. That should fit nicely on an FPGA and, once the bugs are out, on an ASIC. The 1st ASIC is expensive; the 100,000th is cheap.

The StorageMojo take
Flash drives don’t need absolute price parity to win against high-end FC drives. Getting within 30% should do it for most people. Their performance advantages are worth at least that.

Of course the drive vendors aren’t going to sit still. They can pull several levers before breaking the glass for the big red one labeled “margin.” Many have claimed disks are dead and they’re all gone.

But this looks serious. High-end drives are a small piece by units, but their high margins would be sorely missed.

Comments welcome. This isn’t about notebook disks which are currently less than $0.40/GB and headed down much faster than FC and SAS drives.

This is StorageMojo’s 500th post! Thank you, thank you.

{ 11 comments… read them below or add one }

Nik Simpson Monday, 19 May, 2008 at 7:27 pm

Agree 100%, already planned on saying much the same in a presentation next, “Will disks be just a memory?” :-)

xfer_rdy Monday, 19 May, 2008 at 10:09 pm

Yeah, disks are dead, tape is dead and when my grandchildern are retiring they will be hearing the same stories.

Media will stream, and spin and random access for hundreds of years to come. I’m waiting for the return of bubble memory.

When will trade rags stop regurgitating sound bytes like disk is dead. And when will companies like EMC going to start talking about meaningful value for customers instead of selling the same old tripe ? It smelled bad 5 years ago, it’s not doing any better today.

Robin, you are right about comparing raw and finished technologies. They must think everyone is stupid.

Glad to see you called bulls!!t

Sergey Monday, 19 May, 2008 at 10:15 pm

Wait, what about the HDD’s price decrease by 2010?

mrg Tuesday, 20 May, 2008 at 12:38 am

good info -thanks. Can you amend with a discussion wrt watts? Most shops seem to either be ailing from some wattage concern or at least searching for a green story. Do you think this will impact this trend?

Edgar Tuesday, 20 May, 2008 at 12:59 am

Dear StorageMojo,

First of all, let me congratulate you with 500th post.

What I can not understand with current flash drive hype, is that, as far as I know, there is still physical limitation on number of Erease-Write cycles each block can handle. For example, this is even mentioned on Wikipedia: Another limitation is that flash memory has a finite number of erase-write cycles (most commercially available flash products are guaranteed to withstand 100,000 write-erase-cycles for block 0, and no guarantees for other blocks)

Isn’t it going to be a big problem, that could stop flash drives from entering enterprise systems? I don’t think that many people will want to store important data on drives that are “wearing out”, especially for write intensive applications.

Andrey Kuzmin Tuesday, 20 May, 2008 at 2:22 am

> STEC isn’t selling its 74 GB flash drives for $8/GB – $80/GB is closer to the mark.
IIRC, FusionIO claims $2400 for 80Gb board, or $30/GB. No contradiction here?

Wes Felter Tuesday, 20 May, 2008 at 11:34 am

Most of the commenters are missing an important point: high-end 10K/15K RPM disks are dead. AFAIK these disks aren’t getting cheaper, so in 2010 they’ll still be $1-2/GB. Low-end disks have a nice future ahead of them.

When talking about high-end disks, I’m not sure $/GB is the right metric. For many people, $/IOPS is the right metric, and there the crossover is much earlier — maybe it already happened.

Flash has the best IOPS/W and MB/s/W, but low-end disks will have the best GB/W for the near future.

I heard that STEC drives are more like $200/GB, but whatever. Fusion io has a different architecture that shifts some of the costs so that they are not included in the sticker price.

Bill Todd Tuesday, 20 May, 2008 at 6:29 pm

Flash won’t be replacing high-end drives in 2010, because SATA drives will already largely have replaced them (using more spindles and more replication to provide similar performance and greater availability at lower cost). Flash will, however, serve as an important front-end stable write-back cache for disk farms that require high write IOPS (write-aggregating architectures like ZFS’s will take care of the back-end load: the flash front end will be more to reduce average write latency).

Data storage requirements will still be rapidly increasing, and most of the increase will comprise relatively cold and/or bulk sequential data – both of which are served just as well by SATA drives as by the high-priced spread (and at under 1/10th the cost per GB at the raw disk level). Given this, spreading out the hot data across all the spindles required to hold the cold data will in most environments reduce loads to something SATA can handle without breaking a sweat – especially with read-caching and, when necessary, write-caching front ends.

So instead of today’s situation where high-end disks are seeking their little hearts out using optimized queues that do provide the necessary IOPS but only by increasing request latencies to more than the nominal access time of an unqueued SATA drive, most read requests that aren’t satisfied in the front-end cache will target an idle or near-idle SATA drive – and only applications which have real difficulty tolerating the 10 – 15 ms. access time for such uncached requests will need to ask themselves whether paying well over an order of magnitude more to keep all their data on flash is really worth it (especially when they’ll also have the option of just increasing the size of their front-end cache to obtain significant even if not complete relief).

- bill

Barry Whyte Wednesday, 21 May, 2008 at 5:41 pm

IOPs/GB ….

Thats whats interesting.

15KRPM will be the first to go. SAS/SATA high capacity lowprice will be here for a long time to come, archive storage, and tape is still cheaper from a W/YR basis

5yr replacement is an issue, until the $/GB comes down to close to 15K $/GB its all a niche market, but after that who knows…

Joe Wednesday, 21 May, 2008 at 8:53 pm

Fusion IO’s $30/GB is high. Look instead at the Samsung and OCZ 64GB SATAII devices. More like $15/GB.

Anyway, a simple plot of the $/GB trends of both flash and “enterprise class” FC/SAS drives shows an intersection in the 2011-2012 period conservatively. By then, the disk manufacturers will, of course react. The dumb regression is likely going to be, but the disk manufacturers will not take that lying down. High end hybrid drives perhaps? Deeper perpendicular recording perhaps?

The poster that said $/IOPS: flash is already there by a large margin.

There are other factors. How about $/GB per MB/s? I.e., there is a balance between price per unit of storage and price per unit of I/O.

What one needs to keep an eye on is that flash is increasing in b/w far faster than spinning media. We may never see a 20K drive. Conversely, you can order 600MB/s flash cards if you care to right now, and it won’t be long before flash form factor “drives” are all stressing out the SAS and SATA standards.

Donatelli is likely right, but what will come out of the market in response to these pressures is anyone’s guess.

BTW, having done the regression, to the poster who observed SATA will be around for a while: definitely. The regression says that any possible intersection of flash and SATA class drives is 6+ years out. Since that many years is liking speaking in geological time with technology, it would just be really silly to even predict that said intersection will ever happen.

Finally, I think you’ll start to see a tsunami of transparent adoption of flash in enterprise storage. Devices like FusionIO’s make for really awesome journal devices, dontcha know, and architecture’s like Compellent’s will give you a flash breakfast that is really delicious.

Fare thee well, all,

Joe Kraska
San Diego CA

Tony Thursday, 22 May, 2008 at 8:59 am

Well, there are some other possible problems on the way to the flash feast:
1. Flash finally does hit the predicted scaling wall (like the HDD guys, so far they’ve been able to keep scaling, but some of the potential tricks are pretty scary, e.g. 4 bits/cell).
2. Enough flash companies go out of business that it becomes a Samsung/Toshiba duopoly, who can drop prices at a much more profitable rate. This is very possible – go look at the recent results for Micron, Hynix, Qimonda. Heck, even Samsung’s memory business has been hurt.
3. The rising cost of fabs reducing the price drop. Also very possible. The shift from 200mm to 300mm wafers has greatly benefited Intel, Samsung, and the other memory makers. I’ll bet real money that the shift the 450mm wafers will never happen, because the only companies that would benefit are Intel and Samsung. The cost of advanced lithography is skyrocketing.

In any case, yes, I think the HDD companies will have to adjust their product strategies.

Leave a Comment

Previous post:

Next post: