Flash is the least compelling storage technology in ESG survey

Terri McClure at ESG recently blogged about the what technologies are most interesting to NAS customers, according to a survey they conducted last September and October. People following the storage blogosphere may have seen references to Terri's post in blogs written by Sunshine Mugrabi from Ocarina and and Stephen Foskett from Nirvanix

3PAR is not a NAS company, so the survey wasn't a perfect fit, but the results of the survey pertain to more than NAS companies.  There was one data point that I thought was particularly interesting - flash SSDs ranked at the bottom of all the potential technologies people might buy. Instead, survey respondents were far more interested in power, cooling and space efficiencies - in general, doing more with less.

I think Flash SSDs will become more popular over time after the prices become more reasonable, but until then customers are going to continue to look first for products that help them cut costs, as opposed to adding significant costs.  Most customers today are looking for ways to meet their performance objectives while reducing their costs. That's what thin provisioning, wide striped arrays from 3PAR are designed for.

Comments

[Chuck Hollis]

Hi Marc

I guess I'm surprised that you're surprised by this finding.

As we both know, enterprise flash's greatest strengths are high performance random IOPS -- not exactly what most people would use NAS for.

Ummm .. that would be a SAN application, right?

Trying to paint a 3PAR-friendly picture that people just aren't interested in enterprise flash yet flies contrary to (a) most major storage vendors have announced support, (b) a bunch of new suppliers have jumped in the game, and (c) STEC's stock price is through the roof!

You get 2 out of 3 on the standard competitor response to any new technology that you don't have (1) it's not ready yet, (2) you don't really want it, and (3) we'll have it soon. just wait.

Back in 2008 when HP was trying to respond to flash drives, they had some poor chap who said all three things in the same press interview.

I saved it as a masterful example of why you don't put some people in front of the press.

-- Chuck

[Tom Trainer]

Most users I speak with are concerned with doing more with less, as you state. Many are concerned with overall TCO, and some are highly interested/using in SSD. This tends to fit with the survey results. I do wonder if the price of SSD will ever get to the tipping point...?

[marc farley]

Chuck, I'm surprised you didn't chide me for sounding like a broken record because my take on flash SSDs all along has been that flash SSDs are expensive and that people aren't looking to spend more money on storage while the economy is the pits.

But, you might be right about the patterns that competitors fall into because I've also said that the software to use them as something besides LUNs needs to be developed. I guess that's about the same thing as saying they aren't ready yet.

As for "you don't really want it", I haven't really gone there because I think flash SSDs are going to be a big deal once they become more affordable. BTW, that line reminds me of Tim Meadows' lines from "Walk Hard" when he tells Dewey Cox why he shouldn't try various drugs.

[marc farley]

Tom, I'm sure SSD costs will come down sufficiently to make them affordable and reach a tipping point. What price point is that? I don't know.

I think there are also some concerns about long term reliability and maintenance. Those concerns will either be born out or dispelled over time and different SSD suppliers will have different track records. As you know, disk drive reliability follows a bathtub curve with most failures occurring in early life or near the end of life. Who knows what sort of failure curve will be a best fit for flash SSDs.

Just as Chuck identified the ways vendors react to new competitive products in his comment above, the companies that introduce new technology products tend to say things like: 1) It's the future and you don't want to be left behind. 2) Reliability is excellent. 3) It's easy to use and manage.

Of course, the most important point is whether the products are affordable.

[the storage anarchist]

Here's my broken record, which you still have chosen not to respond to:

Flash drives are the most cost-effective way to attain 1ms or faster response times for sequential or random I/O applications. The only other solution is lots of DRAM, but that is significantly more expensive.

While it is possible to attain similar number of concurrent I/Os from a large pool of wide-striped disk drives, there is no HDD configuration that can deliver the per-I/O response times of an enterprise-class SSD. Thus, customers are rapidly deploying SSDs-in volume.

Given the typical protocol overheads of NAS, its not surprising that customers in that market wouldn't see Flash as "compelling" - they probably won't see nearly as much benefit as with block-storage applications. And indeed, the NAS customer's bigger challenges are in dealing with the environmental inefficiencies of most NAS platforms.

And by the way - STEC SSD reliability is only being questioned by those without knowledge of the drive's architecture and availability management. EMC customers have been running these drives more than a year now, demonstrating a MTBPR that is easily better than the best HDDs. You'll come to understand this isn't an issue, once 3PAR actually starts deploying the drives.

Meanwhile, Chuck pretty much nailed it: 3PAR's locked arms around trees #1 and #2.

Meanwhile, every other array vendor does thin provisioning, wide striping, automated storage allocation - and more. When there's nothing significantly unique left in your offering, I guess all you can do is argue implementation...

[Bob Tenda]

Why would anyone waste money putting SSD in a NAS system? That makes no sense. I also find it funny that companies that don't have a certain technology are always finding ways to justify their delay to market. When 4Gb FC came out EMC said "You're not even using the full potential of 2Gb; why do you need 4Gb?" That is, until THEY came out with 4Gb. Then it was a must have feature.
3Par is the only mid-range storage company without SSD so of course they will look for "proof points" to justify their delay. seagate is doing the same thing. They are delaying SSD until 2010 and losing market share because of it.

[marc farley]

So who should we trust about flash SSDs on Celerra, EMC's backpedaling bloggers or their May 2009 best practices planning guide for Celerra, which states how terrific EFDs are for EMC NAS? http://www.emc.com/collateral/hardware/white-papers/h6265-achieving-storage-efficiency-celerra-wp.pdf . As anarchist has blogged about, that's doing the old flash dance!

(note - the above link was bad when it was initially posted but is corrected now)

If that's not enough double speak from the "just say anything" twins, I'll remember that phrase "all you can do is argue implementation" - because we all know that implementation doesn't matter as long you can get "me too" stamped on your data sheet. Here's the Celerra data sheet which includes EFDs: http://www.emc.com/collateral/hardware/specification-sheet/h6035-celerra-ns960-ss.pdf

[Jason]

I find that I agree with Marc (For the most part).

Today everyone wants to do more with less, while I see SSD implementations useful at the very top end of the scale it's just not compelling from a price/gb capex standpoint, at least in the environments/circles I talk with people in.

Also trying to use SSD with some of type block level intelligence in the future just feels like a gateway drug currently to try and get that high margin disk into customers hands, with the 'first hit is free' mentality.

There *are* very interesting SSD implementations that go beyond treating SSD disk/blocks as big dumb storage. The SUN 7000 series with their ZIL/L2arc implementations are very compelling to anyone who has ran them side by side traditional FC arrays. While not a slam dunk in every environment, I think that anyone who downplays their block/nas level abilities needs to get access to one and actually see how game changing moving from a block level array into this type of architecture can be.

Or better yet check out what REAL customers have said after putting them into production.

http://mysqlconf.blip.tv/file/2037101/

(Get's interesting at ~ 19:00 in the video)

Anarchist - Wow, I've got EMC/3par/HP/SUN arrays in my data center right now. Let me tell you from *HARD* experience that just because all these companies have check boxes next to these features in no way means anything in the real world. Their implementations of these features are *VERY* different. The term that best fits here is 'baked in VS bolted on'. And believe me, I'm a fanboy of no array vendor, they all have pros/cons.

I wont even get into cost and maintenance of said arrays over a 5-8 year life.. :)