Blog Viewer


By phansen posted 09-12-2017 01:33 PM


Is the whole industry being misled again, blinded by hype from big firms? ....
75% of all NVMe over Fabric adaptor units sold will be Ethernet-based in the next 2-3 years ... Really?!?  Can you say FCoE déjà vu!?

In an early morning session at the 2017 Flash Memory Summit (FMS) this Summer, a panel discussed the adaptor forecast for the emerging NVMe over Fabrics storage technology, which is primed to revolutionize the modern storage architecture over the next few years. The Market Research session panel members included G2M, IDC, Facebook and Cisco.

All panel members were strong proponents of the NVMe over Ethernet Fabrics solutions (RoCEV2, iWARP, IB), while Cisco did give a meaningful nod to NVMe over FC (FC-NVMe) as viable alternative for certain workloads. G2M provided an industry prospective that by 2020 (2-3 years from now), 75% of all NVMe over Fabric adaptor units sold will be Ethernet-based.

I was shocked!

I stood up, and asked the room of 100-ish people something like; “what single compelling event or reason will change the buying behavior of modern data center storage teams such that they will choose RoCEV2 or iWARP, when for the last decade they could have picked FCoE for their strategic infrastructure and did not? I cannot connect the dots – someone please help me?”
There was NO RESPONSE, just a bunch of shuffling from the panel.

Shuffling because the dots don’t connect. The logic just doesn’t follow when you think it through and ignore the marketing hype like we did on Fibre Channel over Ethernet (FCoE) 10 years ago. The same FCoE that was a total industry failure, capturing no strategic foothold in really any data center.



FC WINS ALREADY AND IS NVME READY, SO IT WILL CONTINUE: Gartner and IDC agree that a vast majority (70-80%) of All Flash Arrays (AFAs) are deployed on a FC network, mostly Gen 5 FC and recently the new Gen 6 FC / NVMe-ready storage networks. Given it’s a FC Fabric, one must assume this business critical asset is managed by the FC or SAN team. Or as I prefer: the Storage Team – the people who know and understand the network functions necessary to support the uniqueness of storage traffic.

And when I think of the applications and workloads running on these emerging, next generation, ultra-fast, NVMe Fabric-attached storage arrays, I see them deployed for those business-critical applications / services that reflect your companies’ best attempt at the optimal digital customer experience, where more and more “customer personalization” from a faster analytics engine gives your company the upper hand in the burgeoning digital economy.

But I don’t see any pending market event that will change the purchasing behavior when deploying these crucial next gen storage assets. Like the prior AFA generation of storage arrays, I believe 75% of these assets will be handed to the Storage Team to be deployed on the highest performing, most secure storage transport in the planet: FC. The same Storage Team that, for the past 10 years, could have deployed FCoE replacing FC as their strategic storage infrastructure - but did not.

STORAGE TEAMS WILL STICK WITH WHAT THEY KNOW: And when one looks out just 2 to 3 years, this same Storage Team will be challenged with which incompatible variant of the NVMe over Ethernet standard will the market adopt? RoCEV2, iWARP, IB, TCP?

Like the Betamax versus VHS wars, this silly skirmish will only make the extremely risk-adverse storage decision maker hesitant to go down any Ethernet-based path until the marketplace and trusted-advisors actually pick a winner in this technology sector.

But as these same Storage Teams deploy their current generation AFAs today, a majority choose a modern Gen 6 FC storage network to go along with storage arrays, as they understand these networks and adaptors are NVMe-ready and fully compatible. Not only can FC and NVMe traffic flow concurrently on the same link (for easy migration), but the raw enormity of over two decade’s worth of technical development of sophisticated FC storage services and functions, automation and analytics, redundancy and security … all the things the Storage Team knows and loves … is available to them essentially on day one since NVMe runs over the existing FC transport they have today.

Cisco admits, what killed FCoE was the “lack of the management layer for storage services”, the things the storage teams requires in order to do their tough job. This crushing gap doesn’t seem to get significantly better with RoCE or iWARP.

THE POINT: Top market research firms are reporting that over the next 2-3 years … with offerings of NVMe over Ethernet versus NVMe over FC Fabric alternatives… Storage teams will choose an Ethernet variant for their storage adaptor, 3 of 4 times.

I think the numbers are exactly opposite. 75% will go FC-NVMe.

And worse, the remaining 25% will be split among the variants: RoCEV2, iWARP and IB. One could argue the IB adaptors will have more early success as the HOC teams pick up the ultra-fast apps., but the other variants … small share. Certainly NOT the 75% share reported by the analysts.


Let’s face it, the RoCEV2 trajectory makes it feel like FCoE V2, with the failure about to happen all over again. I know history repeats itself, but you would think the industry would know better. Guess not.