When is Software Defined Storage not Software Defined Storage?

Software Defined Storage has been a buzzword in the industry now for quite some time. The idea that you can have the smarts of a storage device moved out of a giant datacentre monolith costing hundreds of thousands of $£€ is surely an amazing leap forward for the customer. The hardware producers are maybe not so sure…

Let’s start by defining Software Defined Storage:

Software Defined Storage is the separation of storage software from storage hardware.

Well, duh. What does that actually mean for the customer? Simply, it’s the ability to run the underlying technologies of vendors’ storage devices, should they be Enterprise Disk Arrays, Virtual Tape Libraries, Archive Appliances, etc and supply your own hardware.

Maybe you want to supply one company’s disk array and use another company’s replication and snapshot technology? Maybe you want to have a commodity storage-dense server to supply your storage, but with Enterprise features. Possibly your company has a standardised server platform and you wish to run this, without mixing competitor hardware.

Here’s what has vendors spooked – Maybe you want to break the hardware lock-in which means that a storage vendor supplies your hardware and software. Possibly you want your supplier to see a credible risk that you would move to another hardware provider, if they drop quality, increase costs, deliver new technologies late, etc.

And this brings me on to defining what Software Defined Storage isn’t:

Software Defined Storage is not limiting Features, Functionality or Scale if you don’t buy your supplier’s hardware.

There are various tricks employed by storage suppliers, some blatant, some subtle:

From the data protection world, on the blatant side you may see a company who limits their deduplication disk systems to (let’s say) 64TB, if you want to dedupe more, you need to buy their (commodity, Intel, Linux) appliances. Clearly if a commodity, Intel, Linux appliance can dedupe 100s of TB of disk, there is no reason to limit the disk presented to your commodity, Intel server, running the same OS.

On the more subtle side, you may see a company who allow you to run their deduplication system in Virtual Machines, on various hypervisors – that’s SDS, surely? However, on closer inspection of expected throughput figures and the inability to run on bare metal prevents the system from competing with the vendor’s hardware bottom line.

There are similar tricks for the primary storage vendor: You may find a company who are happy to virtualise third party disk arrays, offering all their top enterprise functionality, replication, snaps, clones, management, etc. The only slight snag is that you have to purchase one of the vendor’s top line disk arrays, in order to hook up a third party storage. More subtly, you may find that a company produces a wide variety of software defined systems, you just have to run them on a hypervisor, the vendor of which that company owns…

Since the recent release of Spectrum Virtualise as software, IBM Spectrum Storage requires no IBM hardware. This is how it should be with all vendors.