Olivier Le Moal - Fotolia
A majority of hyper-converged infrastructure deployments start with three to four HCI nodes that support relatively basic workloads. However, when an organization decides to add more workloads, especially storage-intensive ones, problems can arise.
At the heart of the problem is how HCI typically scales three resources -- compute, storage and networking -- in lock-step even though those resources rarely need to scale at the same pace. Eventually, one or two of them become significantly underutilized.
More efficient use of HCI resources is critical to engender widespread enterprise adoption. Using HCI for only a few workloads and having other more storage-intensive workloads running on isolated hypervisor clusters or bare-metal systems essentially eliminates the value proposition of HCI. Organizations adopting the technology should aim for the opposite outcome, centralizing all -- or all but a few -- workloads in the HCI environment.
HCI environments consist of a cluster of physical servers, called nodes. HCI nodes run on the same hypervisor -- their own or an off-the-shelf variant like VMware vSphere or Microsoft Hyper-V. In most cases, the primary value from an HCI vendor is the addition of storage software and preconfigured hardware, although most hypervisors now provide their own storage software.
Fixed HCI node capacities lead to cluster sprawl
The storage software aggregates the internal storage inside each HCI node in the cluster to create a virtual storage pool or replicates data between multiple HCI nodes. The goal in either case is to provide protection from media failure and enable virtual machine mobility.
These environments expand by adding nodes to the cluster. In most cases, each HCI node is delivered at full capacity. HCI vendors claim customers want it that way, but the more likely reason is that their software cannot integrate capacity added to an existing node. As a result, HCI nodes are delivered with fixed capacity amounts. Some HCI vendors offer high- and low-capacity nodes, but very few have nodes that can expand after implementation.
A small-sized HCI node is the most common one found in HCI clusters, as it enables enterprises to use capacity more efficiently. This approach often leads to node sprawl, which typically means an organization has to add a high percentage of nodes to the cluster to meet capacity demands while, unfortunately, barely touching on available compute capabilities.
Scale-in HCI nodes, and you'll scale HCI right
Instead of deploying a dozen or more small nodes, an enterprise is better served with three or four high-capacity, high-performance HCI nodes that have the ability to add CPU, storage capacity and networking to each node as resource needs increase. The idea is to buy a minimally configured node and expand it instead of buying multiple, fully configured nodes.
Implementing fewer, larger nodes -- that enterprises can expand as needs warrant -- results in a much simpler to manage HCI environment that can handle a greater mix of workloads. For example, storage-intensive workloads that used to be relegated to bare-metal machines can now be integrated into the HCI environment.
The more powerful nodes, meanwhile, have the ability to drive NVMe flash drives and the internal networking required to make those drives perform to their full potential. The fewer, larger node strategy also saves money because it has fewer network connections to make, less CPU resources are wasted and, ultimately, less data center floor space is consumed.
A hyper-converged infrastructure built with smaller, less-expensive HCI nodes seems like a good idea until the node count extends beyond five or six. Then, node management and networking costs start to detract from the HCI ROI.
While the upfront investment for larger, higher-quality nodes may be greater, this configuration enables denser environments that run more storage-intensive applications. In the end, the larger node strategy enables an enterprise to consolidate the majority of its workloads into an HCI environment while reducing networking cost and complexity.