This article is more than 1 year old

Hyperconvergence 101: More than a neatly packaged box of tricks

Simplicity is good, but what else can you do?


VDI is far easier to set up on a hyperconverged box than on a traditional infrastructure or even a converged one, adds Abbott. Whereas administrators had to assign the CPU, RAM capacity and IOPs in the past, these can now be managed by the software, which can be pre-tuned for a VDI environment, he explains.

VDI is also an application more easily served by a single appliance. A hyperconverged cluster can be dropped into a data centre and made to serve a group of desktops without relying on other servers or applications.

This drop-in functionality is the same thing that fueled another key application for hyperconvergence: remote and branch offices. An IT team can install a local appliance with a small number of nodes and scale as necessary.

These two applications may have kickstarted hyperconvergence as a concept, but it’s gathering steam, say analysts.

Fichera points to other applications as hyperconvergence matures, including databases and general line of business applications. “It’s turning into a general platform for any virtualised workload,” he says, but adds that this won’t happen overnight.

Dev-test environments and Splunk-style systems monitoring utilities are also great fodder for hyperconverged kit, claims Prabu Rambadran, senior director of product and technical marketing for Nutanix. But he’s looking further afield. “Think of the rest of the data centre,” he says.

“If you think about virtualisation management, backup and DR, these are other things that people rely on,” he adds, pointing out Nutanix solutions in these areas. He sees the product category fleshing out a layer that’s below the application and above physical networking.

IDC’s Fischer sees another market beyond the enterprise. “Although this is still very much in its infancy IDC has seen service providers show interest in this solution as it offers a very interesting alternative to converged systems when considering capex vs opex,” he said.

The commodity nature of the hardware makes it easier to forklift this technology into high-volume environments, and the ability to easily expand these systems with little administrative overhead could be a big win here. However, hyperconvergence may have to overcome some typecasting as a technology for niche applications before it can make inroads in this space, he adds.

Hyperconverged equipment shows promise in some key application areas based on application design. It is a ‘scale’ out architecture, where each node has its own memory, storage and compute architecture. This the opposite of ‘scale up’, where shared memory and other resources leads to an expensive architecture with complex system bus and node management technologies.

This scale-out model fits well with post-relational No-SQL data architectures. They are well-suited for this kind of hardware because they shy away from ACID transactions, which are designed for centralized software architectures. They are designed to divide up work easily, making them ideal for collections of independently operating compute nodes.

Freeing up the SAN-huggers

Hadoop in particular is a shoo-in for systems with collections of nodes that can handle compute jobs separately to each other, meaning that big data and analytics is a strong use case for hyperconvergenced boxes.

Companies should expect some challenges with hyperconverged systems though, ranging from the political to the technical. Larger companies with existing storage subsystems may experience internal resistance, because storage admins and SAN-huggers may feel threatened.

On the technical side, there will always be legacy equipment to consider. Connecting these boxes with your existing silos will be difficult, warns Abbott. That’s why for the time being hyperconverged kit lends itself to discrete tasks and environments where it doesn’t have to play well with others. Still, IT teams will want to fold it into a broader systems management environment, which could create challenges.

What should IT teams interested in hyperconvergence be doing to evaluate and prepare for these systems? For many, this is still a fresh and untested technology so it will be important to choose the right application and test it thoroughly.

“Pick the right workloads. Choose the required data services and protection, and challenge the HCIS vendor to deliver these services as reliably as a SAN is able to do,” says Gartner’s Butler. Perhaps reflecting the relative immaturity of hyperconvergence as a sector, he warns customers to get a written guarantee that the technology will be sold for the next three years, and supported for at least five.

“Demand proof that different generations of node can be added and supported,” he added.

Look to the future

Customers should also avoid chugging too much vendor Kool-Aid, he warns. “Vendors will be keen to penetrate any data center. They will push near-limitless scaling and no restrictions to uptime or workload choice.”

Make sure these vendors can demonstrate that performance, he advises, adding that when it comes to implementation, customers should test the water with a small proof of concept workload before gradually adding further workloads as confidence builds.

While customers continue to weight up the pros and cons of hyperconverged equipment, the vendors aren’t standing still. They continue to develop the technology in a bid to make it more palatable to a broader audience.

Look out for software-defined networking to complement software-defined storage, say experts. This will make it possible for the hypervisor to allocate storage resources more easily between different compute nodes as the workload demands. If your analytics application is grinding through lack of resource, then SDN could make it easier to throw some flash storage at it, for example.

Hyperconverged equipment won’t be right for everyone, and today customers are still cherry-picking applications where it has proven itself a good fit. Its challenge in the future will be leaping the chasm to broader applications, especially in larger environments where converged systems may be more established and trusted.

The bottom line for those interested? Put your cynicism to one side, but keep some healthy scepticism handy. A little evaluation goes a long way. ®

More about

More about

More about


Send us news

Other stories you might like