NVMe’s predicted ascendancy clouded by architectural hurdles

More than 70 vendors will be involved in the NVMe flash market by 2020 and the market will be worth $57 billion. Meanwhile, nearly 40% of all-flash arrays will be based on NVMe drives by 2020.

That’s according to research company G2M, which has predicted a compound annual growth rate for NVMe-based products of 95% per annum between 2015 and 2020.

But how accurate can those figures be?

The research predicts NVMe will make inroads across servers and storage hardware but also as storage networking equipment to speed NVMe across the likes of Fibre Channel and Ethernet (in so-called NVMe-over-fabrics).

The G2M report reckons more than 50% of enterprise servers will have NVMe bays by 2020, while that will be the case for 60% of enterprise storage appliances.

Meanwhile, it predicts “nearly 40%” of all-flash arrays will be NVMe-based by 2020.

There’s an interesting distinction here that reflects the current difficulties of realising the potential of NVMe flash. Namely, “NVMe bays” on the one hand and “NVMe-based arrays” on the other.

Because NVMe so radically re-works storage transport protocols it is currently held back by storage controller architectures, so cannot realise NVMe’s potential of tens or hundreds of x better performance than current SAS and SATA-based flash drives.

NVMe can slot in and realise its full potential as direct-attached storage – lending it to use in server and storage “NVME bays” – and some vendors have delivered what are effectively direct-attached arrays that lack features such as provisioning, RAID configuration, replication etc.

But the storage controller that must handle protocols, provisioning and more advanced storage functionality currently forms a bottleneck to NVMe use in storage arrays and so we are yet to see a true NVMe flash array hit the market.

So, when the G2M report predicts 40% of flash arrays being NVMe-based by 2020 it it very well as best guess. A guess that hedges it bets on perhaps, but not certainly, that some arrays will have cracked the NVMe-controller bottleneck, and/or that there will also be a number of products that run NVMe in less-than-optimum architectures and that there will also be some “arrays” that are effectively banks of direct-attached storage.

Nevertheless, the research is interesting and shows the perceptions that exist around the potential for NVMe flash.