The Transition to NVMe: Accessibility Matters

Much has been said about the transformational benefits of flash storage. Over the years, I have spoken with many IT leaders on the benefits that their organizations have received after deploying flash storage. Over that time, my position has evolved from thinking it is a great technology to firmly believing that if you are not leveraging flash storage in at least some capacity, you are doing yourself and your organization a disservice. The benefits of flash technology are simply too transformative. Applications run faster and smoother, the rest of the data center becomes more efficient, service time is decreased, and resiliency is improved. The net takeaway is flash storage is good for business.

As a student of system design, however, I understand that flash storage in its current form is limited. The storage media is just one element in the data path. When people talk about replacing mechanical spinning disk with flash, I often hear the phrase, “removing the bottleneck.” That is inaccurate; you can’t ever remove a bottleneck. All you can do is move it. When you deploy flash, typically the storage is no longer the bottleneck, but where did it go? The answer is somewhere else in the data path. To realize the full potential of flash, storage systems and the rest of the components of the data path have to change.

Suggesting traditional storage has to change is not a new idea. The writing has been on the wall for a while now. One step in delivering this new world is NVMe (Non-volatile memory express). NVMe technology is an interface specification designed specially for communicating to flash storage. Older protocols introduce inefficiencies and add latency when communicating with flash. NVMe is designed to solve that.

The storage industry is poised to see a tectonic shift to NVMe over the next few years. Some solutions will inevitably attempt to leverage NVMe to communicate data all the way from the host to the all-flash storage array. The NVMe over Fabrics initiative is an example of one attempt to solve this challenge. If there is one thing that consistently holds true about technology adoption, however, it is that we want the benefits but are often unwilling to change to get them. And if a major change is necessary, such as changing the existing network protocols, then the benefits better be astronomical and even then the shift may take time.

As adoption of new technology ramps, solutions that are the most easily accessible to existing organizations and infrastructure designs often see the greatest traction. Last week, Pure Storage announced its FlashArray //X. The system represents a redesign of the traditional all-flash array, and NVMe is a core part of the redesign. It is funny to use the term “traditional” in front of “all-flash array,” but that is where we are. The bottom line is that the inside of the array is optimized around flash, and I expect, any of the intelligence that once was designed for spinning disk is thrown out. The net result is a faster and more efficient all-flash storage system that fits into its same delivery, support, and management models and is part of Pure’s Evergreen program.

Now Pure Storage is by no means the only player with an NVMe solution, or with one in the works. My assumption would be that every flash storage player is either offering NVMe in some form or has it on the roadmap. But I wanted to talk about the solution, because it is a case study in this idea that new technologies need to be easily consumable.

I am sure there will be some NVMe purists that point out that a full NVMe solution, one that delivers the NVMe protocol, all the way from the server to the storage and back, would be more efficient and faster. While that might make logical sense, IT organizations are wisely reluctant to throw away their existing infrastructure investment to switch to the next shiny new thing. Historically, solutions that require a full redesign will take longer to see adoption than one that offers a familiar deployment model. The NVMe-based flash storage space will be interesting to watch over the next few years. I expect the solutions that offer the easiest path to accessing the benefits will win out, the ones that lead customers on an incremental journey, rather than the ones that require radical upfront changes. But who knows? It will be fun to find out.


Topics: Storage