I had a chance to spend a few days at the Flash Memory Summit in Santa Clara this year, and this blog highlights some of the recent announcements in the AFA space from the show. NVMe was a major theme of the show, and we are seeing more enterprise storage vendors announce NVMe-based features, products and roadmaps.
At the Flash Memory Summit at the Santa Clara Convention Center this year, NVMe technology was a mainstream theme. IDC research indicates that 48% of enterprises already have NVMe deployed in some manner in their IT shops, but 99%+ of this is as local storage that was purchased after market and configured into PCIe slots on commodity x86 servers. While there were several rack scale flash vendors at the show (Apeiron Data Systems, E8 Storage, Excelero), the rack scale flash market is still an emerging market, with only a handful of vendors offering such systems. Two of these vendors (E8, Excelero) won best of show awards this year.
In IDC's All Flash Array (AFA) taxonomy, we recognized that the AFA market fragmented into three distinct segments in 2016: primary flash, big data flash, and rack scale flash. Primary flash arrays are targeted mostly at this point for the consolidation of structured, 3rd Platform Computing workloads, and as such are generally designed for low latency with mixed, small block workloads and include a good set of enterprise-class data services (in-line data reduction, snapshots, encryption, QoS, replication, etc.). Big data flash arrays are targeted at unstructured workloads, and optimized much more for high storage density and low $/GB cost rather than extremely low latency. Both of these types of systems are generally built around SCSI technology, although there are examples of limited use of PCIe/NVMe technology within some of the offerings. Rack scale flash systems are at this point targeted for extremely high performance database and real-time big data analytics environments, and feature NVMe technology end-to-end (NVMe SSDs, NVMe array backplanes and controllers, and NVMe over Fabric host connections). It's IDC's view that NVMe will supplant SCSI as the foundation technology around which enterprise arrays are built in the 2020 time frame. At that point more than 50% of the revenues generated by AFAs will be generated by systems built around NVMe that include no SCSI, and rack scale flash will cease to be a separate segment.
There were interesting announcements in the AFA space among vendors leveraging NVMe. E8 introduced the E8-X24, the first rack scale flash system that leverages dual ported Intel Optane SSDs, and introduced support for shared writable volumes, allowing them to go after Oracle RAC and other parallel environments with their storage platforms. Excelero was recognized as the Most Innovative Flash Memory Technology in the software-defined storage category at the Flash Memory Summit. Excelero's NVMesh software is primarily sold through partners that combine it with hardware and sell it as a storage appliance. One of Excelero's more notable partners is Micron, who uses NVMesh on their recently announced Micron SolidScale, their entry in the growing rack scale flash space.
Kaminario, an established primary flash array vendor, introduced the Kaminario K2.N, their NVMe-infused storage platform, allowing them to pursue both primary flash and rack scale flash opportunities with a single platform. Note that a bit earlier in the year, Pure Storage had also announced an NVMe-infused storage platform, the FlashArray//X, although that system will not support an NVMe over Fabric (NVMf) host connection until 1H18. What was more interesting about Kaminario's announcement, however, was Kaminario Flex, a software-based orchestration platform that makes their K2.N storage platforms dynamically composable across racks within a single data center. Flex allows the on-demand creation of Virtual Private Arrays that draw controller and flash device resources from a common pool, all of which is interconnected through NVMe over Ethernet so that VPAs can be created from resources that span racks, if customers want to. Composability provides extreme flexibility in configuring IT infrastructure resources to meet very specific requirements and in concept is not limited by physical boundaries, but it is in its infancy as far as enterprise infrastructure that includes compute, memory, storage and networking resources. My colleague Ashish Nadkarni has written extensively on this topic, and while the promise of a fully composable, disaggregated infrastructure is still a ways off, solutions like Kaminario's Flex are bringing the flexible composability benefits to storage infrastructure.
Finally, on a more social note, the Flash Memory Summit this year was on fire… literally. On the morning of the show's opening, a fire broke out in the exhibit hall which set off the sprinkler system. The start of the show was delayed about an hour, but the exhibit hall never opened. In one major section of the hall, many of the storage systems that had been set up the night before in booths sustained water damage. Vendors of course were not happy since the lion's share of lead generation at shows like the Flash Memory Summit come from booth traffic, but end users also suffered because they could not roam the show floor checking out the new offerings from vendors. Still, the keynotes, sessions and panels were unaffected and provided a lot of good information and interactive discussion (as well as some inflamed rhetoric) about new developments in flash technology.