Non-volatile memory express, NVMe, has been around for a while; development of the interface standard started in 2007 and it was first released in 2011. NVMe promises boosts in storage performance and much lower latency for flash drives, but the real rewards will come down the road when the interface is paired with the next-generation storage media called storage class memory, or SCM. That’s when data storage will take a significant leap forward.
Like all new technologies, the evolution of NVMe has come in phases. About two years ago, a lot of the big storage vendors began using NVMe as an interface for cache. At the time, it offered a high-speed connection, but was still quite expensive, and therefore was focused only on narrow use cases in the array.
Over the next two years, there’s going to be a land grab of sorts and NVMe will be a given in enterprise storage, just like flash is today. The difference in benefits will be based around implementation. We learned a few years ago with flash that data services matter and proprietary doesn’t work. More businesses require data services to meet requirements for security, protection and availability for their workloads. And a proprietary approach impedes agility and scalability. To keep pace with customer demands, successful vendors will embrace industry standards.
Those lessons will need to be heeded as NVMe becomes more mainstream. It’s also important to understand that in the larger picture, NVMe is only part of the story. NVMe is just the interface or protocol, not the media type. The transitions in interface and media are moving on parallel tracks. On the interface side, NVMe takes advantage of the parallelism in CPUs and SSDs, leaving behind the overhead of storage protocols like SAS and SATA that were designed for spinning disks.
On the media side, NVMe opens the door to next-generation media. SCM is just beginning to enter the picture in enterprise storage and some day may completely replace SSDs. For now, NVMe will mostly be leveraged with SSDs (NAND flash media), which will improve latency, but come at a premium price. That said, SCM like Intel Optane could be the X-factor in the next generation of storage, with much lower latency than NAND flash.
As SCM becomes available in mainstream enterprise arrays, the expense will make it a subset of the overall persistent storage, with the rest of the array being flash. Therefore, it will be critical to have intelligent software built into the array to make cost-effective use of this media. Then enterprises will be able to consolidate all mission-critical workloads onto a single array; you won’t want to have a dedicated array for your high-performance applications and a separate one for the rest of your tier 1 apps.
All-flash arrays are important, but most people in the high end are already there and wondering what’s coming next and what they have to do to future-proof their investment. NVMe will offer a marked improvement in performance and latency over SAS and SATA for all-flash environments, but it will be the pairing of NVMe with SCM that will propel the industry forward.