Modern IT equipment can handle more workloads in a smaller footprint, but this benefit also creates challenges for some enterprises.
A hyperscale cloud provider with full knowledge of its average workload can easily architect a dense compute platform. This is especially true when that average workload is actually millions of different workloads across a massive user base -- which is the case for cloud providers like Amazon Web Services (AWS) and Microsoft Azure -- or is a predictable set of workloads, such as those that run at Netflix, Facebook or Twitter. A single, logical platform that uses a massive amount of compute, storage and network nodes is fairly easy to create for these providers, since it's a cookie-cutter approach; when AWS needs to add extra resources, there is very little systems-architecting involved.
However, it is more challenging for an organization with its own dedicated IT platform to support high-density data centers. For example, an organization won't usually run thousands of servers as a single, logical platform that supports all workloads. Instead, there will most likely be a one-application-per-server or cluster model, virtualized servers that carry one or more workloads and private clouds that carry different workloads with dynamic resource sharing.
Fortunately, hyper-converged infrastructure (HCI) provides a way to better support high-density data centers.
Evaluate HCI -- but carefully
HCI vendors engineer server, storage and networking components to work together and offer adequate cooling at the lowest cost, which enables organizations to support high-density data centers in a shorter period of time. However, there are still challenges with power, cooling and workload capabilities.
Power and cooling challenges are fairly easy to address. Standard power distribution systems can support most HCI systems in a data center facility. But if you want to build a platform that supports high-performance computing (HPC), where power densities might exceed existing distribution capabilities, you'll face concerns. Decide whether expanded power and cooling capabilities in the facility are a worthwhile investment or if a colocation facility can meet these new demands.
It's more difficult to address complex workload capabilities. If you have applications that are directly applied to a platform, hard partition the resources allocated to them, and like in traditional IT models, carefully plan to allow enough space for peak workloads.
When you work with applications that are applied to VMs, remember that each VM is a self-contained entity that carries a full stack of resource-hungry services. Containers share a lot of the same services as a VM, so allow for a greater number of containers to run on a given platform rather than VMs that carry out the same function.
So, just how many VMs or containers should you run on a given HCI platform? Be wary of figures given out by vendors, as the workloads they use to gather those numbers are often generalized and basic. For example, HCI vendors that sell a system focused on virtual desktop infrastructure might state that upwards of 200 desktops can run on their system. But that might only be true when desktops don't have more than one OS and when users don't need to log into them at the same time every day.
Many factors come into play when it's time to choose an HCI vendor. Here are some important guidelines to keep in mind.
Look for vendors who run HCI systems as a proof of concept, allowing you to put your own workloads onto their platform and apply synthetic loads to gauge how many real-world VMs or containers the platform can take.
If you choose to build your own highly dense platform, employ experienced systems architects who can ensure that the interdependencies among compute, storage and network resources are carefully balanced and work well together. People with such skills are difficult to find, though -- another reason why, outside of HPC, HCI is a better bet for high-density data centers than the build-it-yourself approach.
Blade servers offer flexibility, density to enterprise IT
Compare three approaches for scaling an HCI appliance
How do you know if you need HCI?