What I like most about the FlashArray//m is the combination of very dull things that combine to make this an interesting piece of infrastructure.
I tweeted out a cheeky picture of a 6509 with a Pure Storage logo on it, because that’s what the FlashArray//m reminds me of: a modular chassis with a backplane that was the workhorse core switch at multiple clients for over a decade. I quite liked them. We’ve had modular chassis like this for years in networking and server gear, so it’s somewhat astounding that storage doesn’t do this, at least, not in the same ubiquitous way (software based things on x86 servers notwithstanding).
You can swap out modules and upgrade them, and the storage system will stay online the whole time. You can move from one array to a new one, all online. This means you can do rolling upgrades without have to take a massive outage to all of your applications. It’s one of the biggest advantages to software abstractions of hardware, and why vMotion (and Storage vMotion) is so useful.
Honestly, this alone is a much bigger cost saving compared than some assumption based de-duplication ratio. The weeks of negotiating with business units for outage times, the planning, the teams of people doing weekend overtime. Not doing one of those is worth the sticker price of a new array.
It’s also becoming more and more important as storage densities improve, because you end up with more and more applications on the same hardware. Failure domains are important, but for scheduled maintenance reasons more than the infrequent hardware failures, particularly when flash fails less often than spinning disk.
Hardware Defined Storage
It’s also clear that hardware is important, as I’ve said before. Software gives you flexibility, hardware gives you speed, and software ultimately has to run on some sort of hardware. Software provides a lot of benefits, but a good match between software and hardware provides benefits over and above software or hardware alone.
You can get incredible improvements in performance by tuning inner loops in software, but sometimes it’s cheaper to simply upgrade the hardware. But then the performance capabilities of the new hardware are consumed by new features. Anyone who has tried running iOS 7 on an older model phone will know this to be true.
It goes in waves; first some hardware improvements, then some software ones, then hardware again. No amount of software changes can make 100Mb Ethernet go as fast as 10GigE. But who wants to swap out an array just to add snapshots or dedupe?
Pure have matched this technical capability with the business side of things as well, removing various financial and project hurdles in order to keep you buying their storage. Contract renewal, or hardware EOL, are easy times to swap vendors, particularly if you’re staring at a large CapEx bill you have to get approved by the CFO. Smoothing out the spend makes the infrastructure manager’s job a lot easier, and it smooths out the revenues for Pure’s accountants, and keeps the lifetime customer value up.
It’s a nice bit of all-round alignment, and it makes solid business and strategic sense.
Pure’s marketing is getting better, and they’re starting to understand who they are a lot better than they did.
There is still an unhealthy focus on EMC’s XtremIO, but without an appropriately strong marketing message of differentiation. Pure are different, but they’re struggling to define exactly how.
There is still far too much emphasis on cost-based measures. TCO, ROI, payback periods, this is all the language of finance as seen from the infrastructure manager perspective. As I’ve tried to explain before, I don’t buy flash because it’s cheap. Flash is fast. It’s cheap (sort of) when compared to an equivalent amount of spinning disk to achieve a similar outcome, and then I get a bunch of other benefits from flash as well. It’s like saying “I could do this with several tonnes of ball-bearings and a greased sled, or I could just buy a car from you, and the car has a stereo.” Those other benefits help to sweeten the deal, but it’s not cheapness that starts me looking at flash in the first place.
Yes, it takes less power. That might be a motivator if I’ve hit the kVA limits in my datacentre. Same goes for space. In each case, it’s the opportunity cost that I care about, not the direct cash costs. The upside value of the opportunity I’m missing out on is why I buy flash.
Flash lets me do things I couldn’t do before. It could be “deploy more stuff in the datacentre” because now my storage takes up less space, power, and cooling. Yay! But it’s more likely to be “I can do four times the analysis runs I could before” in my data analytics system. Or, more prosaicly, “my VDI implementation doesn’t suck now!” or perhaps “the database performance is keeping the traders happy”.
Pure are still selling into the storage manager, or the infrastructure manager, which explains why their messaging looks the way it does. They’re also selling mostly single-workload solutions, which suits the architecture, and gets Pure a beach-head in the customer from which to expand. It’s a sound selling strategy.
The software itself doesn’t support QoS or real multi-tenancy features yet, and not being able to span across controllers in a scale-out fashion does limit its appeal as an “internal butt” style platform without some major design on the abstraction layer above it. It’s solidly a storage component that a customer might do something else with, and that’s where the involvement of Pure would end and the systems integrators or internal teams would take over.
While this isn’t a ‘disruptive’ or ‘revolutionary’ product, or even especially novel, it looks like a solidly good product with a clear set of benefits, and it should sell well if it performs as advertised.