Last week I met Pure storage at Storage Field Day 6. I was expecting some news, which didn’t happen actually, but I came home with the certainty (once again) that they are the ones to beat if you want to play in the AFA next generation storage space.
I’m not talking about technology, there are many examples of good (sometimes better?) AFA arrays today, but they have the right mindset and the right ideas to play a central role in this market.
At the end of the day, it’s not about every vendor, when it comes to comparisons, having some FUD (or distorted reality) to throw at Pure. But let me explain my point…
AFA: a category that shouldn’t exist!
I’ve already expressed my thoughts in a recent article (and I like to quote myself)
Contrary to general (and Gartner?) belief, considering All Flash Arrays as a separate category is total nonsense. Flash memory is only a media and storage should always be categorized by its characteristics, features and functionalities. For example, I could build a USB-key-based array at home, it is AFA after all… but would you dare save your primary data on it? and what about the speed? (you don’t have to answer, of course!)
The fact that a vendor uses Flash, Disks, RAM or a combination of them to deliver __ promises is only a consequence of designing choices, and we have to look at the architecture (both hardware/software) as a whole to understand its real world positioning. Resiliency, availability, data services, performance, scalability, predictability, power consumption and so on, are the characteristics that have to be considered when evaluating whether an array is good for one job or another.
Who cares about 1 Gazillion IOPS anymore?
Pure plays in real life scenarios, not 4K (or less) block sizes, not 0.1 mS latencies, not Read only workloads, not 1 Million IOPS from a single system. If your (single?) workload has those requisites Pure has little, or nothing, to say. They don’t wear lab coats and they aren’t in the science researching business.
But, if you have performance problems in a real-life mixed workload environment, with mixed and realistic block (it’s likely even your DB no longer uses 4KB, and if so I’m sure that some sort of prefetch mechanism does much bigger reads! Right?), virtualized and physical servers, and so on… they probably have much more to say… and you’ll probably like their solutions.
Don’t get me wrong but, at the end of the day, Pure does nothing special, but it does it right! A block storage with a very good set of functionalities, an easy to use UI, APIs, integration with hypervisors and cloud management platforms, replication, good analytics, and so on. All the features that enterprises find appealing and are willing to buy.
It’s playing in the market of the next generation general purpose enterprise arrays, like many others.
As I said, more than a VNX
A few months ago (another quote by me here), I described Pure FA arrays as resembling EMC VNXs under many aspects… and this has to be seen as an advantage IMHO.
The architecture is similar (Dual controller, scale-up design, primarily a block storage), targets similar markets and has a similar form factor. But thanks to the clever usage of MLC Flash, it’s incredibly faster than a VNX-like (aka traditional) array. Faster enough to be a good choice even to swap out VMAX boxes in the low/mid end.
No scale-out? Yes, federation!
Today, a Pure FA array can scale up to 200/250TB and 200/250K IOPS. Not enough for you? well, buy one more box!
The boxes are totally different entities and they are managed separately, but Pure is working on it.
Today, they already have a tool to manage several arrays from a single interface. Tomorrow, they will be able to manage different arrays as a whole in a federated fashion. (you can find more about this in the video recorded during the event)
Thanks to storage analytics, some automations, and (probably) NPIV, they will be able to automagically move workloads between different arrays to meet SLA, QoS requirements, etc.
Yet again, it’s not scale-out but it will probably be appealing enough for enterprises to buy. This approach maintains most of the advantages you can find in a scale-out system minus some of the complexity. Of course it also has drawbacks like, for example, reaction time to workload changes, cost of data movements in terms of used resources, real QoS very difficult to implement. But, you know, I don’t think most of the cons will be so apparent to average enterprises and end users.
Why it is important
Pure is a well-funded company that is quickly expanding and doing it well (more than 800 systems installed to date). Their product is not the best in the class in terms of scalability or performance? Yes, but it surely is one of the best when it comes to real life scenarios… the types of scenarios encountered every day by real enterprises. Growing sales and happy customers are the proof, aren’t they?
If they can realize their “scale-out-by-federation” functionality, there will be no reason not to evaluate them in the most complex and high demanding environments (I mean high end T1 storage systems like EMC VMAX and the like).
Disclaimer: I was invited to this event by the GestaltIT and they paid for travel and accommodation, I have not been compensated for my time and am not obliged to blog. Furthermore, the content is not reviewed, approved or published by any other person than the Juku team.
Enrico
You could look at Pure’s move as being very clever; they create a product that looks a lot like today’s platforms (dual controller, expandable by shelves etc) that is easy to understand by their target market – EMC high end VMAX. Their offering isn’t the fastest, but is faster than VMAX and is more predictable than VMAX. The result? Customers looking to fix issues with performance on VMAX move over to Pure and get that next step up they need in terms of performance. This makes the sale nice and simple.
The issue of course is how they address the market in 3-5 years’ time, when flash is everywhere and performance demands have moved on from high end VMAX to something more demanding and by which time scale-out not scale-up storage is a preferred model.
So, what’s in the pipeline for development becomes much more interesting to me than the current technology, which was interesting 3 years ago.
I think the company has the sense not to do a “NetApp” and think one product platform can rule everything, so hopefully we will see the announcement of new hardware, perhaps in 2015. Otherwise, raising that warchest of money will have been for nothing.
Chris
Chris,
Good point. I agree, they need to grow fast to become really relevant, but then they have to show more to stay relevant in the long term.
Finally someone says exactly what is needed. As a customer evaluating Pure not because I need 1.363xe^13 IOPS not because I need consultant 95% are and 99.999% < 2ms) because I need to sold a sad tedious story that has gone on for years storage is the bottle neck of all systems today. Processors have kept moving and memory has been forced to keep up but the same 15k disk has to do more today than ever.
Our testing shows that Pure allow our servers to keep being feed with any data they request and ensure users have access to all they need and we can adopt any new business initiative instantly without any excuse or budget issues.
The Gartner AFA category should not exist or Pure should not be in it as it's not the market they aim for. Their drive is replace all disk based storage systems and tier 3 will come soon as SSD sizes increase.
As for if this is the wrong platform vs converged or scale out. I agree with them again why is the question being asked? I'm not sure I would put multi PBs behind a single domain and scale out tends to be driven from a data placement and management perspective. Yes I am sure there are times when scale out is the only option. As for converged I don't quite get the market is simultaneously promoting further separation of parts with things like rack scale architecture and pushing converged. When I look at converged the push comes from simplified management. I know I don't always need compute, memory and storage every time I scale up.