The industry has been crazed by the notion of “Unifying” storage. NetApp has lead the charge as has EMC and others. Above I’ve offered the classic “spork” as a symbol (similar to swiss army knife) of the ultimate flexible eating utility. What could beat it? The problem is I’ve never seen kitchen drawers with sporks. Huh.
At NetApp Insight 2012 the keynote focused on their scale out unified storage capabilities (clustered ontap) and how if fit into the shared storage infrastructure. (BTW, their CTO was pretty good) For the first time I had seen NetApp also accounted for the E-Series, which is a block only array that offers great performance or capacity and is a their fit for Big Data. (BTW congrats to NetApp for announcing their all flash ES540) So the graphic went something like this;
The diagram signified a shared infrastructure approach which has become the default design and architecture view of the modern, virtualized data center. You’ll also notice a breakout of non-shared infrastructure. The intent is to show a requirements driven design for dedicated or purpose built infrastructure I would also call this “specialized”. So why then do I pose the viewpoint of diverging from unified storage? After all the players that have unified all point out how flawed other vendors because they don’t have ALL their features. Most features wins right?
Enter my point. This builds directly off the NetApp diagram above and continues through Chad Sakac’s prediction from his 2013 predictions blog post. (he also verbalized it in his webex) The notion is not purpose built but more like “specialized”. As we are poised to enter a new era of automation at scale not just of size but of “the power to do things” we need to question where we make assumptions. The classic assumption is to reach for flexibility but I’m not so sure. I’ve been a long time advocate for dedupe. What I’ve witnessed has reminded me the lesson of the difference between theoretical and real life results. Theory says you can get x:1. The problem is most IT organization do not and cannot manage the platforms effectively so you end up with a similar gap we have with mileage posted on the showroom and what you actually get when you drive/own the car.
Specialized Storage does not knock a unified platform but instead focuses on how it would be designed/tuned it to the environment. More importantly this opens up a wider variety of storage platforms to have a great fit. To extend that thought you’d want to consider why someone would want an all flash array. I’d offer storage solutions like Nutanix, Nimble, Tintri, FusionIO, Isilon and others that are in the ramp up stages. None of these have the complete solution but each has their niche. That’s good and bad. Good because they can be specialized like EMC and NetApp are even in their unified offerings but bad because you have too many choices. The hardest part is getting them to stop rotating over their product and get tot he business of understanding the specialized fit.
The interesting part here is you’d think I was advocating against NetApp and EMC. That’s not the case. Design objectives will usually lead you to a good “specialized fit”. Here’s some of my supporting thoughts around the idea;
- SDDC – I believe the new era of the Software Defined Data Center will drive a more isolated relationship much the way vVols are being designed (by objective and with a purpose)
- Rate of growth of unstructured content (& Hadoop) far exceeds block storage in a lot of organizations
- The types of growth in data is becoming highly un-compressible (or deduped)
- Storage automation will be driven by non storage array means
Sometimes you need to take 1 step back to take 2 forward. Again, the simple concept of a vVol and where a SDDC total architecture drives us seems to me to be a bit like the automakers after they perfected the assembly lines. At Stratos, where I reside, we’ll be looking into this as we see growth in all kinds of storage. We’ll be developing services around them. We’ll shun the open idea intolerant for a thirst to understand where automation and orchestration takes us in storage platforms and beyond. If you would like us to explore this concept let me know and we can schedule a 1:1 meeting. Comments/Questions welcomed.
@keithnorbie virtual ideas 
Disclosure – EMCer here.
Keith – I think you’re right on. While there are, and will continue to be HUGE demands from customers looking for “sporks”/”swiss army knives” (aka “I have a budget of ~$10-100K, and need something that can host VMs, do NAS, act as a backup target/archive, run my Oracle environment, etc) – there’s no question in my mind that the abstraction of SDDC and automation into other underlying layers (hint hint, for this to work, we still need more “cross storage vendor” automation/API/policy abstraction) is an acceleratant for more “built for purpose” storage architectures (of all kinds)
I don’t know the ins and outs of storage technology, so I won’t pretend to and make myself sound foolish. Instead, I’ll focus on what I do know: food.
Sporks are good enough in certain situations. If you’re going camping, for example, it’s much easier to pack one lightweight utensil that you can use to eat your can of beans at night AND your scrambled eggs in the morning. But if you’re sitting down at the dining room table to eat a porterhouse, a spork won’t cut it (literally). Furthermore, the advantages of a spork (versatility, portability) don’t apply in this scenario, because you have the infrastructure (a full kitchen) to support the best tools for the job (a steak knife and metal fork).
Applying this analogy to storage (or any technology, really), it all comes down to your requirements. If you need to perform multiple basic tasks, the spork approach is the way to go. But if you have specific, in-depth needs, you need specific, in-depth tools.