Dog walkers, the San Andreas fault ... and the storage industry

Hot lava welling up in the San Andreas fault between servers and storage

The aftershocks

Looking more closely at these eruptions, at the hottest hardware part of the fault line we see two things: the future of networked primary data storage is going to be written in 3D Flash and NVMe fabric ink, slowly taking over from classic Fiber Channel and iSCSI. Forget FCoE – it was the wrong answer to the wrong question.

The future of server storage is going to be Storage Memory built from 3D NAND and/or FlashDIMMs, and XPoint DIMMs, if XPoint is for real.

Six recent jolts indicate the way things are going:

• Samsung’s 3D TLC NAND offering 256Gbit chips; 512Gbit ones coming and 1Tbit ones after that.
• These have been taken up by HP, Dell, and Kaminario all-flash arrays, dramatically lowering per-GB flash array cost.
• A 16TB 3D NAND SSD demo by Samsung – most SSDs these days top out at the 4TB level.
• 3D XPoint memory from Intel and Micron, and its 1,000x speed increase over flash, and less-than-DRAM cost. Wow! Maybe storage memory is for real.
• SanDisk ULLtraDIMMs and Memory1 flash DIMMS from Diablo Technologies with near-DRAM speed at sub-DRAM cost.
• Non-Volatile Memory Express fabric; a PCIe network extended outside the server box, which is being adopted by Pure Storage, that well-known endorser of fantastical Gartner all-flash array revenue numbers instead of its own, and also by...
• EMC’s DSSD project – building rack-scale flash inside accessing servers’ memory address space.

That’s the key here; non-volatile memory, flash now and possibly XPoint in the future, possibly some other tech such as phase-change memory or HP’s fabled memristors, but non-volatile stuff anyway, treated as part of a server’s memory address space so that the delay-ridden, disk-based IO stack is shunted out of the way.

The result? A server can run more VMs. How many more? Substantially more. Possibly half as many again. Possibly twice as many. It will be so good that there won’t be any going back, and we’ll wonder how we ever did without it.

In fact, the server storage road forward from here looks pretty golden.

Software-defined storage

On the storage side of the fault line, what about software-defined storage, the hottest topic possibly of all?

First of all, software-defined storage to me is basically storage sold without hardware. Any other use of the term should be banned. To say your storage hardware is controlled by software and is thus “software-defined” is trying to bamboozle people with marketing claptrap. It isn’t software-defined, it’s hardware-limited and customers are locked in. Excuse my Brit bluntness here.

Software-defined storage is a very broad church – too broad I think – and software-only storage products have been around for a long time; think DataCore, founded in 1998, with 10,000-plus customers for its ten generations of storage virtualization software.

And think Nexenta; it’s newer, started up in 2005, and used open source Solaris and ZFS code. It’s passed the 6,000 customer count.

There are also many other software-only storage products:

• Virtual SANS like HP’s StoreVirtual and VMware’s VSAN of course,
• Maxta, Nutanix, and others’ hyper-converged appliance software,
• Caringo, HGST/Amplidata and Cloudian’s object storage SW,
• The Hadoop distros,
• PernixData ESXi caching, and
• Atlantis and its VDI and RAM acceleration of VMs.

There’s lots of the stuff; much of it is niche though. Niches mean silos, and silos mean complexity and management overhead and cost.

DataCore and Nexenta are more general storage, with DataCore being, for me, a narrow-band (in marketing terms), SMB-focused offering, whereas Nexenta is broad-spectrum storage, energetically going for the enterprise, with its tech covering primary and secondary data storage, object storage, Big Data, and the cloud, as well as SMB storage needs.

The concept of SW-defined storage, even when restricted to SW-only storage use, is still too broad for practical use. Products like VMware’s VSAN, Cloudian’s RING and NexentaStor are so vastly different in scope that to lump them together as SW-defined storage verges on being ludicrous.

You can sort of understand storage hardware marketing droids jumping on the software-defined storage bandwagon. They have hardware to shift and what better way to pull the competition’s teeth than by dressing their hardware in a software-defined invisibility cloak, and talking about separation of control planes and data planes.

These are important distinctions, and help contrast different forms of software-designed storage, but they apply after we split overall software-defined storage into software-only and hardware-linked.