As we have watched the flash storage industry evolve over the past several years, vendors tend to follow one of two different approaches. The first approach is often one that tries to time the market and figure out when to make the transition. This approach tends to involve tiers of storage infrastructure, where more emergent technologies are more expensive and targeted at higher value workloads, while the established technologies are delivered to the masses. As time goes on, technology vendors figure out how to balance the transition. And we in the industry debate things like the cost benefit tradeoffs of the different options.
Pure Storage seems to be taking a second approach to flash. Instead of focusing on how to balance different offerings and time the market transition, Pure emphasizes accelerating the delivery of new innovations in flash. Pure’s approach pushes the envelope, pushing new flash innovations out quickly while extending its flash technology into an ever-broader set of customers and workloads. One example of this approach was Pure’s launch of FlashBlade, all-flash storage for both file and object data, a couple years ago, back when many in the industry doubted that flash could ever be effective for large capacity unstructured data stores.
Last week at Pure Accelerate, Pure’s FlashArray//X announcement gave us another example of this approach. 2018 is the year of NVMe, and we have seen multiple NVMe related announcements already. The IT execs that I talk with love the idea of NVMe. Their logic goes something like this: flash is already great. NVMe is a protocol designed for flash. It will make flash faster; what is not to love? Everything you love about flash, just better. In ESG’s research into NVMe-based flash storage, 79% of storage decision makers said that they expected NVMe-based flash to replace traditional (SAS/SATA based) flash storage.
And, while some in the industry debate when to make the jump to NVMe, Pure announces NVMe across its portfolio at, according to Pure, a zero-dollar price premium.
Skeptics are quick to call out, “price premium over what?” It’s over the FlashArray//M, obviously, and (continuing my skeptic voice) Pure controls the price of FlashArray//M. But here is the thing, price is often a function of both component costs and competitive forces. If Pure’s decision to go all in on NVMe put them at a significant disadvantage versus the non-NVMe based competition, they likely wouldn’t be doing it. And with the more entry-level FlashArray//X options, Pure is offering non-NVMe based media to help get prices lower, but the NVMe architecture is still inside.
Moving past NVMe, Pure Storage announced its AI-Ready Infrastructure (AIRI) Mini. Go, right now, to any IT vendor website or twitter account and count how many times artificial intelligence is mentioned. Seriously, I’ll wait….
Ok, a lot, right? Everyone is talking about it. AI is the future, it is just a matter of when… Here is the thing, though—building infrastructure for AI is a little different than your normal virtual machine environment. And, GPUs are just the beginning. For AI to take off, the infrastructure that supports it can’t be a science project anymore, it must fit within the paradigm of IT operations. Pure and NVIDIA teamed up and decided to do something about it, rather than waiting for “when” to get here.
Now, I might be reading too much into this and pushing the envelope with technology may not always be the best strategy, but I am bullish on what Pure is doing here. There are certain technologies that are not a matter of whether they will take over, they are a matter of when. Flash is one of these, NVMe is another, and I would argue, along with everyone on Twitter, that AI is another. Instead of waiting for these products to hit some component cost target or some adoption level, it makes more sense to start now, and build the solution you expect the future to need today. This was my takeaway from Pure Accelerate. Pure isn’t waiting for the future to happen, they are laboring to make it happen today.
Now, Pure’s success isn’t guaranteed and there is still a lot to be done. The latency benefits of Pure’s NVMe architecture are impressive, and they are expected to even improve with NVMe over fabric support, but the question remains: will it be enough to deliver the data-centric architecture that Pure is envisioning. One that can replace the DAS and NVMe-based architectures supporting the more emergent cloud-native applications, while also accelerating traditional applications. For that, we will have to wait and see. But given Pure’s aggressive approach, they are well positioned to make their data-centric architecture a reality.