IT vendors are riding the converged infrastructure wave into data centers by labeling their wares with the term,...
but it turns out that very few platforms are truly converged.
To prepare for his session, Converged Infrastructure vs. Commodity Hardware at the upcoming Modern Infrastructure Decisions conference in New York City on April 11, Bob Plankers took a whirlwind tour of the myriad converged infrastructure options currently available.
What he found surprised him: Converged infrastructure means many different things to many different vendors. In this Q&A with Plankers, a senior data center and virtualization architect for a major Midwestern university, he described the basic categories of converged infrastructure, their virtues and their shortcomings.
What surprised you when you surveyed the converged infrastructure field?
Bob Plankers: The wide range of what people are calling converged infrastructure surprised me a little bit. Some organizations use the name 'converged infrastructure' to basically start a conversation with a customer, but there's really nothing converged about it, except maybe the storage.
From what you can tell, what are the different categories of converged infrastructure?
Plankers: At the lowest level, there's SKU-based convergence, in which you order all the discrete pieces of the environment at the same time, and it only looks converged because you basically bought a bundle. Think of it as the IT equivalent of the model boat kit that you bought as a kid. It's got all the little plastic pieces and the model glue, but you still have to put it together yourself.
NetApp FlexPod is an example. They'll sell you a whole pallet of stuff through one of their resellers, but it's basically an opportunity for one of the VARs [value-added resellers] to come in and do a bunch of consulting work. At the end of the day, you're managing the storage array separately; you're managing the virtual infrastructure separately; you're managing the servers separately; or you're using the plug-ins to vCenter that NetApp provides -- stuff that you could have done on your own.
Bob Plankers,senior data center and virtualization architect
Then there's also management-based convergence, where you still have a bunch of discrete components -- disk array, servers and network switches -- but with a common management console. The poster child for that is probably the VCE Vblock, where you buy one Vblock and you know that it will run x-number of VMs [virtual machines]; you just drop it in, and you're done. The unit of infrastructure is large, but for a lot of organizations, that works. Think about what organizations like eBay are doing for servers. Their unit of servers is a whole shipping container full of servers. This is basically the smaller enterprise version of that.
Finally, there's total convergence from vendors like Nutanix. There, the basic unit is a little bit smaller -- a 2U server that you just keep stacking up in your data center, cabling it up with 10 Gigabit Ethernet [GbE] or InfiniBand, and when you need more capacity you just add more. You get CPU, memory, networking, storage capabilities in each unit and then the whole cluster just figures itself out.
What's converged infrastructure's Achilles' heel?
Plankers: It depends on what kind of converged system you're talking about. With management-based convergence, it's the size I think. You end up having a minimum size that you end up having to buy into, and the minimum size tends to be too big for a lot of smaller organizations. That's where the 'Nutanixes' of the world can really shine, because their units are a lot smaller.
With total convergence, where you just add another building block to the cluster, the management interface isn't trying to disguise the fact that it's managing a set of discrete components like with management-based convergence. That can be a good thing and a bad thing. It's great when it works, but when that black box fails, it can be a little tricky to troubleshoot.
But the real Achilles' heel for all these systems is that they don't connect to anything you have, and if [they do], [they are] completely unsupported. With converged infrastructure, you need to be able to treat your existing infrastructure as a sunk cost and walk away from it. That's hard for a lot of organizations, because [the existing infrastructure] just works, and it's doing just fine. Maybe the organization sees the promise of converged infrastructure, but they want to go at a slower pace toward it.
Five years from now, will we see more or less converged infrastructure?
Plankers: I think it will stick around because it's a good idea in a lot of cases. Convergence is a way to compete against the public cloud idea; to gain some cost efficiencies; gain some economies of scale and keep things local where people have a better sense of control and things. Long term, these converged infrastructure units are going to get more robust; the feature sets are going to improve and they're going to become more popular.