A few weeks ago when EMC’s acquisition of XtremIO was finally announced, financial analyst firm ISI characterized EMC as the “Baskin Robbins of Flash,” a moniker that EMC itself then adopted and promoted last week at EMC world (you can watch Pat Gelsinger’s keynote to understand their positioning). When I first heard the Baskin Robbins characterization I thought it must be an insult, so I was a bit surprised to see EMC actually embrace it. Here’s why:

Who Wants 31 Flavors of (Flash) Storage?

For the purposes of this discussion let’s ignore host-based PCIe flash (which is on a path to ubiquity, and solves different problems than networked flash). In the world of networked flash, how many flavors of flash does one need to serve all the market use cases?

For networked deployments, EMC really has four major flavors of flash it is currently juggling: 1) All-Flash Symmetrix VMAX, 2) Symmetrix VMAX with FAST tiering between flash & disk, 3) Project Thunder (40 GbE or InfiniBand-connected flash appliance), and 4) XtremIO (FC-connected all-flash array)…plus a 5th and 6th flavor if you want to count the VNX variants on the above.

EMC’s problem? To serve customers best all these flavors should really be one thing. In the ideal case they’d be able to evolve the Symmetrix product line to be the all-flash platform, but for reasons we’ve covered in depth before flash requires a new architecture…so let’s say that’s just not possible. In this case, they certainly shouldn’t be building two next-generation networked flash platforms. EMC has thus far positioned the difference between Thunder and XtremIO as “server-networked” vs. “storage-networked,” but that distinction is thin at best. Let’s explore at the similarities:

  • Both have commodity flash architectures – ✓
  • Both need RAID to ensure resiliency against device failure – ✓
  • Both need deduplication to somewhat counter-balance cost – ✓
  • Both need copy services features (snapshots, replication, etc) – ✓
  • Both are positioned to do 1M+ IOPS – ✓
  • Both are networked block devices – ✓

So…the major difference seems to be that one is connected via a FC network, and one is connected via a lower-latency IB or 40GbE network. But in this world of multi-protocol storage, can’t one networked all-flash platform just support both connections? Especially when the hardware in both these products is industry-standard commodity components, and the differentiation is all in the flash management software stack.

The reality is that EMC will certainly sort this out…eventually. They are just caught in an awkward moment with two alpha-level products on their hands built by two wholly-separate engineering teams, and it is a bit too early to pick one of the horses to bet on. So, my advice to customers would be to wait until EMC picks a flavor, because if you pick early you’ve got a good chance of picking the wrong platform. And while ice cream flavor choices only have to satisfy you for 10 minutes in the parlor, you live with a storage platform choice for years and years….you don’t want to get stuck with an abandoned cone melting in your data center.

Where’s the Affordable Flavor?

And of course the real shame here is that despite the myriad of flash flavors that EMC is now offering, they are missing the one flavor that customers actually want: an affordable one. As we’ve stated before, EMC is unapologetically performance-focused in its flash strategy, delivering Tier 0 flash solutions that are geared towards high-performance deployments at a high-cost to match (>$30/GB usable). This positioning is natural, they have a legacy disk business that they need to maintain and protect. But with the right recipe, flash can still deliver incredible performance, but do it at the cost profile of existing Tier 1 storage.

Pure Storage: Wholly-Focused on One Flavor

When you add it all up, it turns out that Pure Storage believes in quite a different approach. Our strategy is simple:

  1. Develop a single all-flash storage platform for the enterprise that allows customers to completely migrate away from spinning disk for their performance applications.
  2. Focus on beating Tier 1 spinning disk economics with all-flash, specifically driving the cost of flash down to the $5-10/usable GB range. This allows for Flash to not just live in the land of Tier 0, but to substantially disrupt existing Tier 1 storage.
  3. Deliver multiple protocols from a single platform. We’ve started with Fibre Channel as the obvious place to begin, but you’ll see us add pretty much every other block “wire” that’s out there, as well as file protocols over time.

There’s no confusion or placing multiple bets over here at Pure Storage…we’re 100% focused on building all-flash storage at a price point that enables pervasive deployment in the enterprise data center. If this is the flash flavor you are looking for, drop us a line, we’d love to let you taste a scoop.