Alexandr Mitiuc - Fotolia
Recent years have seen a steady growth in hyper-converged infrastructure adoption. A portion of this growth has been fueled by the fact that hyper-converged platforms have become far more flexible. At one time, deploying HCI meant almost certainly purchasing preconfigured appliances.
Today, however, you can implement hardware-agnostic HCI platforms through software. So users no longer necessarily have to increase compute, storage and networking resources in lock step.
DIY hyper-converged cluster pluses and minuses
Building your own hyper-converged cluster deployment using HCI software has its advantages and disadvantages. Keep in mind the following when implementing a hyper-converged infrastructure deployment.
One advantage of DIY software-based hyper-convergence is you pay only for what you need. HCI appliances generally include compute, network and storage resources. An organization that wants additional storage under the traditional hardware node approach to hyper-convergence might have to buy an entire module rather than just the storage it requires, for example. This means paying for unneeded compute and network resources.
In contrast, software HCI makes it possible to use the hardware configuration that makes the most sense for your enterprise.
Need more storage? Then all you have to do is buy more capacity, no additional network and compute are required. Hardware used must adhere to the hyper-converged software vendor's compatibility list to be fully supported, however.
One of the biggest disadvantages of the software approach to hyper-convergence is HCI software doesn't usually enforce hardware uniformity. Here, organizations must be especially careful to use matching hardware in their hyper-converged clusters, or it can negatively affect the IT environment. Consequences vary depending on just how mismatched the hardware is, and what the HCI cluster deployment is being used for.
How mismatched HCI hardware muddles capacity planning
Mismatched hardware can make capacity planning far more difficult. This is especially true for hyper-converged cluster deployments used to host virtual desktops.
One reason why HCI works well for hosting virtual desktops is hyper-convergence makes it easy to add storage capacity. The virtual desktops within a virtual desktop infrastructure (VDI) group are uniform, so each requires identical system resources.
Because of this, each hyper-converged cluster node within a standardized VDI deployment can host a specific number of virtual desktops. Any time an organization needs to host additional virtual desktops, all it needs to do is determine how many virtual desktops each HCI node can handle. From there, it's a simple matter of extrapolating that number to how many nodes will be needed to acquire the needed additional capacity.
While it's possible to use mismatched hardware for virtual desktop hosting, doing so makes it far more difficult to determine the resources that are needed when extra capacity is required. That's something to keep in mind when considering software-based HCI.
How mismatched HCI hardware complicates load balancing
Another potential consequence of using mismatched hardware in an HCI deployment is it can complicate load balancing for your hyper-converged environment.
If each node contains an identical set of hardware resources, then it is relatively easy for a load balancer to evenly distribute virtual machines (VMs) or applications across the available nodes, and no single node has to handle a disproportionate share of the total workload. If the hyper-converged cluster's hardware resources are mismatched, however, then the load balancer will need to have knowledge of the mismatch, complicating load balancing. If it doesn't have that knowledge, there's a risk of some nodes being over provisioned while others are underutilized.
How mismatched HCI hardware increases chances of failure
Another potential consequence of using mismatched hardware is that -- depending on the software being used and the nature of the discrepancy -- the mismatched hardware could cause load balancing and high availability to fail because hypervisors tend to be sensitive to differences in CPUs. This means hyper-converged clusters based on HCI software may not be able to live migrate a VM to another node if that node is running an older-generation CPU.
There are workarounds -- such as Hyper-V's CPU compatibility mode -- that can enable live migrations between mismatched CPUs, but using these features can negatively affect performance. Furthermore, doing so can't overcome all CPU differences. For instance, it would be impossible to live migrate a VM from an Intel-based node to a node containing an AMD processor.
Implications of mismatched HCI hardware: The Takeaway
Unlike hardware-based HCI, software implementations make it possible to build hyper-converged deployments from mismatched hardware, relieving enterprises from the reliance on fixed resources in the identically or similarly matched hardware of hyper-converged cluster nodes. Using matched hardware makes HCI deployments easier to manage and maintain, however, due to potential complications engendered in areas like capacity planning, load balancing, high availability and migration when using mismatched nodes.
It is important to take the implications of these complications into consideration when investigating which approach to take, hardware or hardware-agnostic software, for your hyper-converged infrastructure.