Rather than getting caught up in the definition of hyper-converged vs. converged infrastructure, some vendors are...
blurring the lines between the two with new product types that take lessons from both CI and HCI.
Since HCI became popular, the name has been applied to many different products that do not conform to the accepted definition of hyper-convergence. There has been a progression from what vendors call "best-of-breed" infrastructure to CI, then to HCI. Best of breed involves separately evaluating servers, network and storage and then assembling the best choices for your environment. Deploying such systems involves a lot of cabling, software installation and custom configuration. It can take weeks or months to stand up a large environment; even ordering the right collection of cables can be a challenge.
In the traditional hyper-converged vs. converged fight, CI puts the selection and integration of the components in the hands of the vendor. The CI product is a fixed collection of hardware that is delivered as a pre-integrated system. Usually, CI systems are available in a small number of fixed sizes: medium, large and extra-large. Each unit contains all of the components and cables but is a single item to order.
In theory, deploying CI is a matter of unboxing and connecting the power and network feeds. A CI system is usually ready to run virtual machines (VMs) within a day or two of delivery. CI simplifies ordering and deployment, but it does little to help with the ongoing operation of the deployed infrastructure. CI systems tend to start with platforms for hundreds of VMs and scale to tens of thousands.
HCI is a customizable system that is designed for running VMs. HCI eliminates the storage network, and the shared storage is integrated with the hypervisor hosts. The hosts form a distributed storage cluster across an Ethernet network. HCI systems are usually quick to deploy; it can take under a day to set up a medium-sized cluster. HCI is also focused on operational simplicity, eliminating storage management and integrating backup management into VM management. This ongoing simplification is a core feature of HCI products. Another core feature is the ability to scale the storage and compute capacity by adding HCI nodes to the cluster. An HCI cluster can start small and grow over time as the workload increases.
New entries change the hyper-converged vs. converged game
A couple of products have blurred the lines of the hyper-converged vs. converged battle, possibly creating something entirely new. In 2016, Datrium launched a disaggregated storage array for virtualization called DVX. The persistent storage is in its disk shelf (called a data node), but it does not have the conventional storage controller to manage RAID and snapshots. The storage controller is software in servers running the VMware ESXi hypervisor. These ESXi servers share the data node and have solid-state storage for the performance tier, which is accessed directly by the controller software on the server. Putting the controller and SSDs in the ESXi servers means that they scale out as the administrator adds more ESXi hosts (and, therefore, more VMs).
In addition to the data nodes, Datrium sells ESXi servers with flash, called compute nodes. Customers can reuse their existing ESXi servers by adding solid-state storage and installing the controller software. Datrium does not call DVX hyper-converged; it uses the term "open convergence." Simplified management without compromised performance is a stated core Datrium principle. This simplification matches up with what hyper-converged is all about, although the Datrium architecture is not hyper-converged.
NetApp, on the other hand, does call its product hyper-converged but doesn't put storage inside the ESXi servers. So, some say, it is not HCI. The NetApp product pairs a SolidFire scale-out storage system with ESXi servers in the same physical chassis. The chassis is a "4 servers in 2U" format, like most Nutanix systems, as well as EMC's VxRail. This way, there is a scale-out storage system alongside a vSphere cluster, all packed into the same high-density server chassis. The benefit over traditional HCI is that the admin can scale storage capacity and performance independently from compute capacity. NetApp provides a very simple deployment; whether or not operations are as simple won't be clear until the product is in the hands of more customers following its general availability release in October 2017.
Both Datrium and NetApp segregate storage from compute but deliver the business value that HCI wants to provide, changing the rules of the hyper-converged vs. converged game. They address limitations of the HCI model where storage is inside the ESXi servers. Expect to see more of these advanced HCI architectures over the next couple of years as more vendors apply the lessons of HCI deployment to product development.