3 Considerations When Evaluating Hyperconverged Infrastructure (HCI) vs. Custom Data Centers

If you’re evaluating storage solutions for your data center, you may be trying to decide between a custom data center build and newer hyperconverged infrastructure (HCI) appliances. I recently had the opportunity to review an HCI appliance, and I was pleasantly surprised by many aspects.

What is HCI?

An HCI appliance is like a server rack or cabinet with everything you need to run a small data center or cloud. It contains one or more firewalls or gateways, one or more network switches, power backup, cooling, a storage solution, and blades for all the compute needs.

The cabinet comes preconfigured and internally cabled, so it’s ready to be rolled into your datacenter, network closet, or your parents’ garage, then plugged into a power source and network drop to provide a fully functioning data center appliance. Some HCI solutions can be chained together to work as one. It often comes pre-installed with storage, network, and node-management software, the blade operating systems, and even container orchestration platforms (such as Red Hat’s OpenShift).

HCI pros and cons

Both HCI appliances and custom data center builds have strengths that make them more suitable for certain environments. If you’re evaluating HCI vs. a custom data center build, the following information may help you decide which makes more sense for your organization and your specific wants, needs, and requirements.

Convenience and speed

HCI’s most obvious benefit is convenience. For many years, I’ve participated in cluster builds, both physical hardware- and virtual machine-based. My team’s last bare-metal build — with redundant firewalls, load balancers, multiple availability zones, high-speed redundant SAN storage, multiple routes in and out, and highly configurable blades — took over a year. This included the build’s conception, through planning, financials, and ordering, and then installation, which included cabling, redundancy cooling, redundant power sources, and everything else you might expect in an enterprise datacenter multi-cage (or multiple failure zone) cluster build.

Contrast that to the HCI environment I recently used, which took a few weeks from start to finish. Once we had approval for the project, the appliance was ordered, delivered, installed, and up and running within three weeks. Our only significant delay was the high-power cabling, which our existing cage was not equipped to handle, so the turnaround time could have been even shorter.

Configurability

You get significant convenience with HCIs, but there is a natural tradeoff with configurability. With an HCI, you don’t get all the configuration freedom you would have with a custom, ground-up build, and you pay more upfront for an HCI appliance. On the other hand, you don’t pay the salaries of highly technical experts you need for a custom build, and you don’t have a long delay before you can use your hardware with an HCI.

The HCI appliance will also be more opinionated than a custom build. This means that the HCI architecture team made some decisions about how your platform is built and configured. For example, with an OpenShift HCI, the architects may have chosen the number and size of the management nodes, the ingress capabilities, which plug-and-play virtual networks are included, the platform version and update schedule, and more.

Whether this is a pro or a con is up to you. If your staff includes experts in these areas, you may not want to be locked into another team’s architectural decisions. However, I suspect most organizations considering an HCI appliance will be happy to have experts making these decisions to make the entire system work well on day one.

Expansion and geographical limits

Another consideration is that each HCI appliance is independent of the others. They cannot span multiple cabinets, but they can be chained together to make multiple clusters work together as one unit.

While planning a data center solution built from HCI appliances, ask how many can be chained together, whether there are geographical limitations, how they behave with workloads when they’re connected, and any other questions specific to your data center needs.

Software and support

The final aspect to consider when evaluating HCIs and custom data centers is how the software and support model works. With HCI appliances that include the necessary management and compute platform software, updates are extremely smooth. Updating the firewall firmware, kernel, drivers, container runtimes, and everything else can (and should) be handled during one window. All the drivers and software have been tested together, and the upgrade and migration path has also been tested. This should provide very smooth and predictable maintenance upgrades.

Contrast this with a custom data center, where you must investigate every device and the full software stack for compatibility, order of operations, data migration, and more. Updates can go very wrong, and it’s much harder to estimate the time to completion.

Support for the HCI appliance is streamlined: You talk to one vendor that is responsible for the complete stack. They will direct your inquiry as needed and work closely with the teams handling support. For example, if you need additional support from the appliance hardware maker, the HCI vendor will handle that interaction and chase down solutions for you.

With a custom data center build, you must contact and coordinate support with each vendor. Sometimes, a vendor might say they don’t support your configuration, which means you have to figure out how to solve the problem yourself.

Conclusion

Hyperconverged infrastructure is neither a baby step into building data centers nor a replacement for custom-built data centers. Weigh the pros and cons as they affect your organization and its requirements. Consider whether you can update node hardware, add GPUs, or expand storage and memory. Research any expansion and chaining limitations and whether you can chain to other geographical locations. Understand the upgrade cycle and the vendor’s decisions around networking, redundancy, and disaster recovery.

You might be thinking, “these are the same questions we ask when building custom data centers”, and you’re right. The problems and decisions are the same, but with an HCI, someone else has made well-educated decisions about them. Above all, HCI is an interesting and convenient new delivery mechanism for modular data centers.

Mike McDonald is an SRE Manager & Senior Cloud Architect at IBM based in Vancouver, Canada. The above article is personal and does not necessarily represent IBM’s positions, strategies or opinions.