VFD4 Prep: SimpliVity

SimpliVity make a series of “hyper-convered” arrays.

The devices are a mix of commodity CPU, RAM, and storage in a single OmniCube with a special bit of custom hardware: a PCIe card dubbed the OmniCube Accelerator. This custom hardware offloads the high-overhead calculations required for things like compression and de-duplication from the main CPUs, freeing them up to serve VMs.

Software Runs on Hardware

It’s a clever combination of software and hardware, recognising something often overlooked by people blinded by “software defined” hype: software runs on hardware.

If you just have clever software, and run it only on general purpose CPUs, those CPUs get more loaded than if you offload specific functions onto custom hardware. This is why these days we have dedicated switches, routers, firewalls, and so on, because back in the day, everything ran on this thing called a “server” connected to other “servers” using network cables. Servers were routers.

Having functions on your network card, like the TCP offload engine, means the central CPU doesn’t have to do all the TCP calculations. Similarly, if you play games, you’ll have a graphics card. Remember maths co-processors? Same thing. There are very real, very valuable reasons for having custom hardware to perform specific types of calculations.

What’s clever about SimpliVity is that it they use commodity components where that makes sense, and just write x86 based software for them. The custom card uses the standard PCIe interface to talk to commodity gear, and SimpliVity get the best of both worlds in a way that’s easy to manufacture.

Simplicity

I like the mantra of simplicity from SimpliVity (and the name is a pun, I get it). Finally we have IT companies understanding that they should do a bunch of hard work to hide all the complexity from their customers. Pick sensible defaults. Make it easy to do the right things. Make the computer do all the hard work.

It should be easy to just move workloads around onto other OmniCubes. Why should it be hard to move data to different physical devices? It should be easy to turn off a bit of kit to replace it with new ones. We’ve been doing hardware upgrades for decades. Why is data migration always such a colossal pain-in-the-ass? Adding a new bit of hardware or compute into the existing pool should be easy.

Anyone would think IT people liked making their jobs difficult and were doing it on purpose!

Available in Cisco UCS

SimpliVity and Cisco announced a partnership at VMworld US 2014, which I think is a pre-cursor to SimpliVity getting acquired by Cisco. Cisco need a bigger storage portfolio now that the VCE alliance is winding down. The Whiptail/Invicta flash stuff is still a bit too new, and all-flash isn’t practical for the majority of enterprise customers; they still need spinning disk. SimpliVity makes a lot of sense as a scale-out bolt-on for UCS, and if they do get acquired, I’d expect to see stand-alone units getting deployed alongside Cisco switch gear into companies doing some sort of refresh.

This partnership is the companies’ way of testing out their relationship. Living together, or going on a long holiday, to make sure that things are likely to work out in the long run.

Cisco will still need a bunch of other things in their portfolio, so I’d expect a bunch more partnerships or acquisitions, if they do decide to become a provider of all your compute, networking, and storage needs. There are still a lot of people buying standalone arrays, and hyper-converged isn’t suitable for everything.

VFD4

I’m looking forward to seeing what SimpliVity have to say from a virtualisation front. This is their first full slot at a Tech Field Day, after appearing at some other Tech Field Day events.