There are more hyper-converged infrastructure vendors today than ever before, but that means there are more definitions of the technology as well. That's why it's important to keep the hyper-converged infrastructure definition simple, George Crump, founder of analyst firm Storage Switzerland, said.
For him, the simple hyper-converged infrastructure definition is: "the convergence of storage software into the virtual infrastructure." For administrators of virtual environments, that means once difficult tasks such as provisioning storage for virtual machines are simplified, while storage features are provided to each node. Hyper-convergence goes one step further than converged infrastructure, which typically pre-integrates storage, networking and compute, but forgoes the storage software component.
Crump added that in today's hyper-converged market, there's plenty of options on the table. Some vendors ship pre-configured nodes, while others provide what Crump described as a "recipe" -- a pre-defined architecture that the administrator must assemble themselves. Other hyper-converged products come in a "pure" form, in Crump's words, which administrators can download in the form of software and integrate into an existing environment.
But hyper-converged architectures are still typically deployed in conjunction with a new project, such as virtual desktop infrastructure or the virtualization of applications like Microsoft SQL or Exchange. There's still work to be done before hyper-converged infrastructures make a splash beyond small- to medium-sized data centers, according to Crump.
"One of the challenges with hyper-converged infrastructures has been guaranteeing performance," he said. "But vendors are now progressing to where they're getting better and better at making sure I can guarantee certain service levels to particular applications."
Transcript - Hyper-converged infrastructure definition grows with vendors
A couple of years ago there were a lot of varying definitions of hyper-converged storage. What is your hyper-converged storage definition, and do you think it's become more standardized today?
George Crump: Well let's take that in reverse order. There's probably more definitions now than there's ever been. Every vendor now is wanting to claim some type of hyper-convergence because it's a buzzword, and they want to jump on the buzzword. From a definition standpoint, what I would say from a really simple level is it should be the convergence of storage software into the virtual infrastructure, or the hypervisor, and that's where we get "hyper-converged" from. Some examples of varying definitions that we see are a turnkey approach to hyper-convergence where everything comes from one vendor. We're seeing what I'd describe as a recipe, or pre-defined infrastructure, but you still go get those parts. And we're seeing the purest instance where you just download software, put it into your existing infrastructure, and now it becomes hyper-converged.
What about converged infrastructure? What would you say is the difference between converged infrastructure and hyper-converged infrastructure?
Crump: That's a really good question. Convergence is essentially the pre-integration of existing components, so we're taking a server, we're taking networking and we're taking storage, and we're putting it into a rack ahead of time, plugging it in, making sure it all works, and then delivering that to the data center. The value to the data center is that all they have to do is plug it into the wall and start creating virtual machines. But we're not fundamentally changing the way things are purchased; we're just eliminating some of the integration steps.
Hyper-convergence does something a little bit different. It takes the storage software that was on the storage system, and moves that into the virtualization layer, and runs either at or near the hypervisor, and that's where we get hyper-convergence from.
Hyper-convergence changes the way we deploy storage, but how does it change the role of the storage manager?
Crump: I think fundamentally the role of the storage manager actually stays somewhat similar, because you still have to manage capacity and manage performance. And the different solutions will give you different capabilities as far as doing that, and that's one of the things that will differentiate solution A from solution B. But fundamentally, the role of the storage manager stays roughly the same. There might be tighter integration into VSAN if it's a VMware environment, or whatever the hypervisor might be, so there might be some automation, but I think the role and importance of the storage manager stays roughly the same.
What would you say the top hyper-convergence use cases you're seeing today are?
Crump: By far the big one is virtual desktop infrastructure, and part of it is the way you buy a hyper-converged infrastructure. Typically what you're looking for is a new project or a new opportunity -- we'd call it a greenfield opportunity -- though VDI is a new project for a lot of people, so they're taking hyper-convergence and applying it to virtual desktops. It scales a little more naturally into that type of environment so it works really well there. We're also seeing it in increasing amounts of back office environments, so Microsoft SQL, Exchange, SharePoint and things like that where a lot of those applications have not been virtualized yet, so it's another greenfield opportunity and you want to really apply the right resources to it.
Just in the past few years we've seen hyper-convergence go from a really niche product to something that a lot of the major vendors are offering today. What do you think the next major development in hyper-convergence is going to be?
Crump: As far as hyper-convergence goes, I think it will be moving up market. Hyper-convergence in its initial form was really good for pet project like virtual desktops or mid-tier data centers. One of the challenges with hyper-converged infrastructures has been guaranteeing performance. Because once you start to converge everything you now share everything and so making sure that application "A" gets "X" type of performance becomes more difficult versus say a dedicated storage infrastructure. The vendors are now progressing to where they're getting better and better at making sure I can guarantee certain service levels to particular applications.