Scale-Out HCI is not Enough

Scale-out is the primary focus of most hyperconverged infrastructures (HCI), but scale-out HCI is not enough to deliver a more cost-effective and easier-to-operate data center. HCI vendors claim to offer scale-out by adding nodes to increase computing power, networking bandwidth, and storage performance/capacity. As VergeIO points out in its latest blog, “Scaling Infrastructure in Three Dimensions,” scale-out infrastructure is not enough. HCI doesn’t scale out very large. It also can’t scale small or scale vertically. 

However, I want to add another scale dimension to the mix, “depth.” Depth of scale ensures that each node within the infrastructure is used to its maximum potential before adding another node. Like the other dimensions of scale, HCI needs to improve depth of scale and is something that Ultraconverged Infrastructure (UCI) possesses almost uniquely. 

What is Deep-Scale Infrastructure?

HCI adds nodes long before the software fully utilizes the existing nodes’ computing power and performance capabilities. An architecture like UCI, which delivers deep-scale, ensures that the investment in the existing nodes is fully realized before the organization spends the time and money adding another node. 

Deep-scale requires two capabilities within the software. First, that software must be efficient. Efficiency is an area where HCI struggles because of its layered form of convergence. With HCI, although they are on the same node, virtualization, networking and storage are three separate layers. 

Because HCI vendors don’t always own the hypervisor, they can’t optimize it. They also can’t run the storage and networking parallel to the hypervisor. Most HCI solutions require that networking and storage run as a virtual machine (VM) within the hypervisor, subjecting them to the virtualization tax. The virtualization tax means that IT professionals can’t utilize the full potential of the processors and storage within each node. They must leave plenty of headroom to pay the tax. The overhead gets worse as the node count increases. 

UCI Delivers Efficiency

UCI integrates virtualization, networking, and storage into a cohesive data center operating system (DCOS). Storage and networking run as equal citizens, not as VMs. The integration negates most of the overhead of virtualization, enabling near-bare-metal performance for virtualized applications. This efficiency enables IT professionals to run more workloads on less physical hardware than the typical HCI solution. 

VergeIO’s DCOS solution, VergeOS, also integrates global inline deduplication. This integration not only reduces the cost of storage investment but does so without adding the typical overhead associated with the deduplication algorithm. VergeOS’s storage capabilities are so powerful that many IT professionals consider it when facing a storage refresh.

Node Flexibility

The second aspect of scaling deep is node flexibility. Most HCI solutions are severely limited by the type of nodes that can exist in a single cluster. Those nodes must have very similar configurations, the same or similar processors, and the same or similar storage media. The similarity means customers needing, for example, just capacity must add processing power. Not only is the original investment underutilized, but now a part of the investment is “double-under utilized.” As the HCI cluster scales, utilization of the hardware declines rapidly.

UCI Delivers Node Disparity

With UCI, nodes of different configurations can be members of the same DCOS. Nodes are grouped by type:

1. Blended Nodes – A balance of CPU and storage performance/capacity.
2. Compute Nodes – A set of nodes with higher-end processors, including GPUs.
3. High-Performance Storage Nodes – A set of nodes with NVMe SSDs.
4. High Capacity Storage Nodes – A set of nodes with high-capacity hard disk drives.

An unlimited number of node types can be organized into clusters, and one DCOS consolidates those clusters consolidating their capabilities into a global resource pool. The global resource pool can provide those capabilities universally across all workloads to specific workloads that can take advantage of them. For example, not every workload in the data center needs a GPU, but those that do typically benefit significantly from its availability. 

At VergeIO, we refer to this node flexibility as part of our scaling vertical capability, but it also fuels our ability to scale-deep. Grouping nodes by their potential and intended use case limits resource burn. 

Combining node flexibility with the upfront efficiency of the software, VergeIO customers will find their percentage of hardware utilization to be much higher than any HCI. A higher utilization percentage directly correlates to a reduction in upfront hardware spend and ongoing operational costs. It also means that as the organization adds new workloads or new, more innovative hardware comes to market, IT planners can graft those workloads and solutions into the existing DCOS instead of creating an entirely new infrastructure. 

To learn more, join VergeIO CEO Yan Ness and me for our live webinar “How to Eliminate the Data Center Scale Problem” TOMORROW, Thursday, January 19th at 1:00 pm ET, 10:00 AM ET.