Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

Hyper-converged system challenges: Performance, integration

Despite the reasons a business might invest in a hyper-converged system -- easy integration, scalable in nature -- the technology comes with drawbacks.

According to George Crump, founder of analyst firm Storage Switzerland, it doesn't help that hyper-converged infrastructure is still over-hyped, and it's important to evaluate options with eyes wide open.

"It's clearly a shared everything environment," Crump said. "So, isolating performance to a particular workload and guaranteeing performance can be a challenge."

Crump said a good rule of thumb to avoid lagging performance is to be sure the hyper-converged system can provide more IOPS than the most an application will need at any given time.

Resource overhead is another area that users should look out for. Due to the nodal structure of a hyper-converged system, capacity and performance often need to be scaled in lock-step, whether both need to be scaled or not. This may not be as noticeable for small-sized businesses, but for larger ones this can introduce some inefficiency. Some hyper-converged system vendors are working to offset this problem by offering nodes that are used specifically for capacity or specifically for compute.

However, large enterprises may run into more hyper-converged system challenges than smaller-sized businesses, which could account for the faster adoption rate of the technology in remote offices and small- to medium-sized businesses. Crump noted that all types of environments get more complicated as they grow, and hyper-convergence is no different.

"As you get to 16 or 20 nodes, you might want to pay more attention to the inner networking between the nodes," he said.    

A final area Crump said to keep an eye on is compatibility. In many cases, a hyper-converged system will not integrate with an existing SAN or NAS. Many hyper-converged options also support only one hypervisor.

"If you want to mix hypervisors, or you want to pick a particular hypervisor, that's a real limitation in some of the hyper-converged architectures," Crump said.

View All Videos

Transcript - Hyper-converged system challenges: Performance, integration

We often hear about hyper-converged benefits like easier management and scalability, but we don't really hear about the challenges often. What would you say some of the challenges are?

George Crump: One of the things about hyper-converged storage is it is over-hyped. If you talk to many hyper-converged vendors they can do no wrong, and that's probably a little optimistic. I like to go into this eyes wide open and look at some of the drawbacks. Number one, it is clearly a shared-everything environment, so isolating performance to a particular workload and guaranteeing performance can be a challenge. Number two, many of them don't support the existing storage infrastructure. So if you have an investment in a shared SAN and you've already made that investment you would probably want to leverage that into your hyper-converged infrastructure. Being able to support that is a key capability to look for and many hyper-converged solutions don't support that. Finally, and maybe most importantly, many infrastructures only support one hypervisor, and in many cases it's their own customized hypervisor, so you're not necessarily running VMware or Hyper-V or whatever hypervisor you want to use. If you want to mix hypervisors, or you want to pick a particular hypervisor, that's a real limitation in some of the hyper-converged architectures.

As you mentioned, one of the big problems with hyper-converged storage is having to scale compute and capacity together. Does that mean users end up with a lot of resource overhead, and how can they cut back on that?

Crump: Absolutely. Another big challenge of the hyper-converged architecture is the requirement in most cases to scale performance and capacity in lock-step. And most data centers don't scale like that. Hyper-converged vendors will try to get around that by selling you essentially a cheap and deep node that has nothing but capacity, for example, but you're still adding some level of compute so it's not perfect. In other cases you might be able to buy a hyper-converged architecture where some of the drive bays aren't fully utilized and you can come back and add drives later, but still it's difficult to perfectly balance the two. As a result of doing all that, people do end up with one resource that's out of balance. They either have way too much capacity, or way too much compute, and there's no rule of thumb, you're either going to be one or the other. In some cases you may decide as a business decision that that's fine because all of the other benefits of hyper-converged work in your environment.

Would you say that there's a sweet spot for the size of an environment that hyper-converged storage works best for?

Crump: As you scale anything, it becomes more complicated. And what we're really talking about in a hyper-converged environment is a highly clustered environment and these nodes all have to talk together, so the more of that we have, the more complex the environment becomes. So you can scale, but you have to be very careful how you scale. Also, as you get to maybe 16 or 20 nodes, you might want to pay more attention to the inner networking between the nodes. So I think for most people, as you get to a certain level the appeal of hyper-convergence starts to dampen a little bit because you just don't have that simplicity that you started with. So to answer your question, typically today if you're in a mid-tier data center, hyper-converged is very compelling, it's very economical, it probably solves most of the performance problems you have. If you do get out of balance in performance and capacity it's probably not going to be extreme. So those markets work really well. That's not to say it doesn't work in the enterprise -- certainly we have hyperscale data centers that say otherwise -- you just need to be more careful, and you need to have better planning and a better IT team to be able to manage that environment.

What are some management features that you think are most beneficial to hyper-converged products?

Crump: I think one of the management features that is most important is [the vendor's] integration of the storage layer into the hypervisor that they're supporting. In some cases, some of the hyper-converged infrastructure works with VMware, so they've written some specific code to plug into vCenter so that provisioning, data protection, resource allocation becomes very seamless, right through the vCenter console. Then there's other guys that have done the exact same thing for Hyper-V. Then the other type of converged infrastructure we see is where they've used a customized version of a Linux hypervisor and they've integrated into that. So that's probably the single most important management gain, is this very tight integration into the hypervisor that's been selected to really make provisioning and assigning VMs easier.

What are your top tips that you would give a storage manager for managing a hyper-converged environment?

Crump: I would say number one is performance -- making sure that the hyper-converged infrastructure at its worst is going to perform better than when your application is going to be its most demanding. So if you have an application that occasionally needs 20,000 IOPS, your hyper-converged infrastructure should always be able to deliver 25,000 IOPS. Then you're fine. But if those are out of balance, it becomes a key problem for you. So that's number one. Number two, look to see if you're in an environment where your capacity and compute resources are totally divergent of each other, because you would be one of those people that would end up with either way too much compute or way too much capacity. So be very cognizant of that. And I think finally, look at the future. One of the things we always look at with hyper-converged infrastructures is how flexible they are. Are they really that flexible? Will they allow you to change to another hypervisor if you decide that makes sense in the future, or will they allow you to integrate existing storage resources if that makes sense? So flexibility and future forecasting become very important.   

+ Show Transcript