The Butterfly Effect for All Flash Arrays

 Butterfly Effect
A chaotic effect created by something seemingly insignificant, the phenomenon whereby a small change in one part of a complex system can have a large effect somewhere else
< At our Insight event in Berlin last week, NetApp made a whole bunch of announcements in front of over 3,500 of our customers, partners and employees. Sitting in the press briefing room reviewing the announcements, we came to a rather innocuous one. The launch of the 3.84TB TLC SSD’s for our All Flash FAS (AFF). Of any announcement, a simple capacity increase for SSD has to be one of the most basic, right? I mean it happens like clockwork a couple of times a year. The group moved on and began discussing all of the other new things but I couldn’t get past this. There was something niggling in my head. I kept asking myself of the new “Only-Flash” Array vendors, which ones are using these drives? Pure? Nope, it’s taken them almost 12 months to go from 1 - 2TB. EMC XtremeIO? Not yet, only just announced support for 1.6TB SSD’s. After a number of talks with a number of colleagues I found out there’s a fundamental reason why. Their myopic view of “only-flash” is now becoming the inhibitor to something as fundamentally simple as adopting a new capacity drive. If you decide to create a single large pool of storage and apply your efficiency globally across this pool then you need to keep a very large amount of meta-data in memory. As you increase the SSD sizes (essentially doubling each time) you need to massively increase the amount of memory inside the storage controllers to deal with this. A good detailed description of the different technical approaches to this has been written by my colleague Dimitris here

For Pure this would appear to be proving a challenge. It means adopting new SSD’s is currently way behind the launch of the new capacities The industry is currently at 3.84TB and they are stuck on 2TB which they only just announced. And when 16TB SSDs become available for Enterprise Arrays, no amount of Global efficiency is going to make up for the fact you have to buy 8 x SSD’s instead of 1.

If you own Pure arrays or are considering them, then you should absolutely be asking for a very comprehensive and accurate road-map on their plans for this.

For XtremeIO it’s even worse. They suffer from the same issue as Pure and have only just announced support for 1.6TB SSDs. Again, when you look toward the future, you should definitely ask them for an accurate road-map for when new capacities will be adopted.

But, (and it’s a HUGE, Gob-smacking, Oh my word, kind of but), if you’ve bought a couple of EMC Xbricks today with 1.6TB drives and you decide to add more Xbricks in the future then they have to be identical. Yes, that’s right, identical. Even if they could introduce a slightly higher capacity disk in the near future, you cannot use it without starting a new cluster.

When I was discussing this with some colleagues I had them repeat it several times as I couldn’t quite believe what I was hearing.

Because of our AFF design choice NetApp can rapidly adapt. 3.84TB SSDs? You got it! A higher capacity next year? No problem.

We are now seeing that the narrow architectural scope from Pure and XtremeIO is rapidly becoming a very significant limitation.

I’m sure some of you reading this are probably thinking, “well he would say that he’s from NetApp’ right?” well yes I am so you know the source, and don’t get me wrong there is some very clever technology in these arrays, but this is one of those significant limitations that’s kind of glossed over in the flurry of the sale that you should absolutely know about up front and not find out 12 – 18 months after you’ve made your AFA purchase.

Talk to us to find out how an architectural vision that spans from flash to disk to cloud expands the incredible performance, efficiency, adaptability and huge range of enterprise features available in our AFF platform. Don’t let your next AFA purchase leave you exposed to the impending ‘butterfly effect’.


  1. NetAppian here.

    Great post, Matt. In an age where flexibility, ability to adapt to change, and time to… absolutely trump a clever design choice, it really has me scratching my head. We know hardware commoditizes over time. We know hardware commoditizes (speed increases; capacity increases) faster than software developers can push out a new release. To design a system based on hardware flat-lining? If the value prop of your system is “we sell the same thing they do for half the price” well, then you pretty much understand that hardware IS the commodity which makes it even more curious that you would design things in such a way that you couldn’t ride commodity density and price curves. Penny-wise, pound foolish.

  2. Great post as ever Matt, clearly highlights the potholes that the flash in the pan vendors are now hitting. Unfortunately their expensive marketing campaigns and the fact they give valuable rewards to successful partner salespeople means that they’re going to shift way more than they deserve to, despite these shortcomings. The interesting point will come when vendors such as NetApp are way ahead on all flash capacity and the Pures of the world can’t explain it away by pointing at their pretty orange bezel. Plenty of low hanging fruit around that time methinks….

  3. Interesting take…however it is obviously biased and you completely “glossed over” several factors that limit the validity of this article: 1. The fact that EMC has an entire portfolio of Flash products and not just XtremIO. If you want an array that you can put large capacity SSD drives in for cheap, then absolutely do so with an All Flash VNX. If you want rich data services, significant reduction in management time, power cooling, longevity of your SSDs, minimal write amplification and all of the benefits that an All Flash Array built from the ground up brings, then go with XtremIO.

    Bottom line, there are use cases for each and only EMC gives you CHOICE when it comes to Flash because of the breadth and depth of their portfolio.

    1. Appreciate you taking the time to comment, I’m not so sure I did ‘gloss over’ several factors, you mention about using VNX where someone would require putting ‘large capacity SSD drives in for cheap’, however I notice that all your VNX spec sheets state that 1.6TB drives are currently the highest capacity, which is exactly the same as for XtremeIO, do you charge less for them when they are installed in a VNX? I also noticed that these are eMLC SSD’s so also pretty unlikely to be a cheap option when much of the industry has moved to TLC for it’s capacity and price benefits.

      My point about being ‘stuck’ with the drives you get with your first Xbrick is also a very relevant and valid architectural consideration for anyone looking at XtremeIO. When its highly likely that we’ll see 16TB SSD’s next year people should know what this means if they have or are considering your product.

  4. Great post! This is totally reflecting the experience I made with NetApp. You can find plenty of such stories where NetApp is doing way better as any other storage vendor but not positioning it. Currently I initiated a workgroup together with some NetApp colleagues regarding security, compliance and privacy. Topics that are currently buzz-words, especially in Europe. Everybody jumped on the “Security-Train” but NetApp? NetApp has all the features matching compliance requirements since years! And as NetApp has it since years, they become to be normal… All the other storage vendors has to apply third party functionality in order to get at least some requirements fulfilled but NetApp has it all integrated. Why has the best technology always the worst marketing? I hope we can change this a bit with our workgroup “NetApp integrated Compliance”.
    Stay as you are and keep going!

    1. Thanks for the comment Marcel, we will keep going, and work harder to make sure that we’re really highlighting the rich features and value we have in our solutions.

  5. I believe Kaminario is the only AFA vendor who uses 3D TLC drives, however the largest capacity they ship today is 1.9 TB. Of course the 3.8 drives already exist (just cram in double the NAND) but their performance characteristics are not exactly the same as the 1.9 TB drives. Nice to see NetApp not resting on their laurels anymore and beginning to lead the pack again. Sometimes all it takes is a glimpse of your own mortality to start livin’ right. 🙂

    1. The 3.84TB drives we are now using are the 3D TLC drives, there’s been a huge amount of work done on optimising read / write paths in ONTAP for SSD so the performance for these is looking extremely good, we’re also seeing very good write endurance. There is definitely a renewed sense of speed at NetApp right now, release cadence for ONTAP is good and every release is bringing great new features, onwards and upwards.

      Thanks for your comment

  6. Tell me if I’m wrong somewhere but you are saying because NetApp doesn’t provide (and will not by the way) dedup or compression on the aggregate basis, we can deliver more capacity storage. This is an obvious statement for anyone not providing better efficiency storage solution but still want to play in the field. Let’s take an example of datastores hosting preprod (integration, dev, qualif, etc) and prod VM where we want snapshots. Efficiently I should put everything in the same volume (dedup scope) but with the snapshot policy, it will not be good. You see the efficiency and the adaptability… You are talking about your core product. To get the same features like Pure, you need to add crypto, the clear and integrated dashboard, 24h/24 monitoring on the web and the fact that storage admins do not have to run any updates on those arrays.

    Of course, NetApp has some pros but you need to work on your cons and it is not only marketing. Others vendors have challenges also. Customers have specific workloads and like never before, in house, really good system admins are the keys to pick the right solution to their specific context. In the era of cloud and IT commoditization, those guys are not worthless. Thanks for the reminder.

    1. Thomas, there is a lot of very interesting technology that we are working on that I cannot comment on in this post or this reply, but your first statement is a bit bold for someone that doesn’t work for NetApp and therefore has little or no knowledge of any plans that we might have now or in the future, but let’s let that pass for now.

      Back to your comment, no that is not exactly what I was saying. I am noting that all design centers have tradeoffs. They are real and act both as inhibitors and accelerators. The real trick is to pick a design center where the accelerators are the ones that provide long term value and have a long life. IMHO, a design center that covers the meat of the market is always the right one. Just this week we’ve seen several commentaries about how slightly higher efficiency may well be completely negated by the ability to rapidly adapt to use higher capacity SSD’s. For your example about Integration, dev, qualif etc then unless you give this some size then it’s a perceived issue without any actual known impact, are these 1TB databases, 5TB, 10TB? and then it comes down to design, why not keep all the supporting DB’s in the same volume so there’s all your efficiency, then keep the prod one in a separate volume and now you have your alternative SnapShot schedule. Is this less efficient? possibly, but if you’ve been able to use much higher density SSD’s with a lower price point then the effects may well have been eliminated.

      We also find a siginificant number of our customers like to use our arrays for many different purposes, NFS for example, you’ll need something else for this if you went for Pure, CIFS, again you’ll need something else for this if you bought Pure. We want to take all of this into account when someone considers a storage strategy, one array can very quickly become way more efficient than having to buy many.

      So back to where I started, we know we have to keep pushing forward with innovation, but like I said all design centers have trade-offs and the real trick is to pick a design center where the accelerators are the ones that provide long term value and have a long life, and this is key to our focus and the discussions that I have with the most important people within the companies we engage with.

      Thanks for contributing to the discussion

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.