As with any technology that quickly catches the attention of the enterprise storage market, there may be more to it than meets the eye. In the case of hyper-converged infrastructure, it's important to remember that the low cost and easy deployment don't mean the technology is a save-all.
For example, hyper-converged technology combines storage and compute into nodes, which provides management benefits. Instead of dedicating separate management to networking, servers and storage, it can all be controlled from a single pane of glass. Nodes eliminate infrastructure complexity as well -- organizations that need more performance or capacity can simply add nodes as they grow. But this also means there is a potential for overhead because compute and capacity need to be added in lock-step. An organization that needs to add capacity may find itself paying the price for more compute than needed.
In this Tech Talk interview, Ben Woo, managing director of analyst firm Neuralytix, discusses the scalability problem and other limitations of hyper-converged technology. According to Woo, in some cases, it could be larger organizations that are more likely to see the limitations. For example, performance can be hard to guarantee for specific applications in hyper-converged environments. "Here's a scenario where you've got the best infrastructure in the world, but if the applications aren't aware of the underlying scaling and all those nodes, it becomes pointless," Woo said.
Woo also pointed out that other concerns expressed by potential hyper-converged technology buyers, such as vendor lock-in, may not be as detrimental as they may first seem. "You might start with one company, but as the applications grow and scale, all you care about is, is there enough compute power, is there network connectivity and is there enough capacity?"
Transcript - Hyper-converged technology hurdles: Performance, lock-in
Do you think that scaling is a limitation for hyper-converged storage?
Ben Woo: No, I would argue that a converged system potentially offers a scale-out limitation. They tend to be a little bit more involved in the deployment. They tend to take up a lot more floor space than hyper-converged. I think what you see is a little flexibility with hyper-converged systems because you can use it in the remote office, and you can use them in the data center, and, in both situations, the scaling component is really what we're talking about. In the remote office, you might have one, maybe two nodes for redundancy, and then your main data center back in headquarters, might scale out to tens, maybe hundreds of nodes.
Easier management is one of the big selling points of hyper-converged technology. So how exactly does it change the management for the storage administrator?
Woo: Well, it's not just about storage, and I want to make sure that we have an understanding here. Hyper-convergence is about bringing together compute, network and storage. How is it different? It's actually quite basic. In the old days, we used to have to manage the compute with a server file. We used to have to look at managing the network discreetly, and we had to manage the storage discreetly. Now it's all in one.
So what that means to the IT infrastructure is we can spend less time optimizing, configuring and provisioning IT services to spend more time on the information or the data side. Ultimately, it's the information that makes money. The infrastructure costs money and the information makes money. So we can spend less time on infrastructure technology, and focus on what IT really stands for, which is information technology. We ultimately get a lot of benefits out of it.
How would you say hyper-converged systems typically do with performance, and do you have any tips for better performance for those systems?
Woo: Performance is the trickiest question out there. What is high performance to you is sluggish for somebody else. Again, we have to categorize what type of company it is. For the most part, medium-sized companies and small companies don't really have a true performance problem, and scalability overcomes a lot of this anyway. You have a performance problem, throw another node in there.
For the large-sized companies, performance is about balancing the load, and that's where applications come into play, and I mentioned a little earlier that application may not be ready yet. Here's this perfect scenario where you've got the best infrastructure in the world, but if the applications aren't aware of the underlying scaling and all of those nodes, it becomes pointless.
What about the cost model of hyper-convergence? How do you evaluate it and is it really cheaper than getting a traditional array?
Woo: You can't just look at Capex. Is the Capex going to be lower? Yes, there's no question about that because you're buying fewer things. The major savings come in the Opex side. So no longer do you need a server administrator, a virtualization administrator, a network administrator and a storage administrator. Somebody can manage this holistically. So, particularly for smaller firms, the IT generalist -- that's how we refer to these people who have multiple skill sets -- can now be in a better position, again, to optimize, configure and provision the IT services necessary.
In a large firm, it's slightly different. You still tend to have specialists in various types of components. Again, storage, server, et cetera, and that actually creates a little bit more tension and potential opportunities for conflict because now everything is the one footprint, and the question comes down to who owns what. But I think that's another question for another day.
What options do people have that want to move to hyper-converged infrastructure, but are concerned about vendor lock-in?
Woo: Hyper-converged infrastructure has the benefit of being somewhat agnostic to software and vice versa. Ultimately, we're driven by application, and the great thing is it doesn't really matter what the infrastructure is. You might start with one company, but as the applications grow and scale, all you care about is, is there enough compute power, is there network connectivity and is there enough capacity at a certain level of performance? So, vendor lock-in actually is gone completely. It does put a lot more pressure on hyper-converged infrastructure vendors, but, ultimately, we're looking at the hypervisor app, and that's really down to whether you use VMware or another form of hypervisor and then whatever software you're running.