The three tiers of hyper-converged architectures

Hyper-convergence is the consolidation of the compute, network and storage layers into a single tier. The goal is to drive out complexity and to drive down costs.

While it may seem that the turnkey nature of hyper-converged architectures absolves an organization of the responsibility of caring how the storage infrastructure works, the opposite is the case. Understanding how the storage software works, what storage and media was selected and why will help the IT planner make sure that the right product is selected.

To better understand how hyper-converged architectures work, it's important to drill down into the three tiers:

The compute tier is made up of two parts: the physical server hardware and the hypervisor software that gives hyper-converged its name. Both elements are critical to the overall success of the hyper-converged project as well as the suitability of a particular hyper-converged architecture for the organization. There are three compute configurations available:

• A bundled hyper-converged system where the software and hardware come pre-integrated, and all future upgrades must come from that vendor. This configuration sacrifices long-term flexibility to gain the simplicity of a turnkey system.
• Software-only configurations where the organization provides its servers. This approach sacrifices simplicity of setup for long-term flexibility.
• A "software-mostly" model. In this case, the software can be installed on almost any server, but it comes with some proprietary hardware such as a PCI Express board to run specific functions.

The hypervisor software will also vary. It can be one of the brand names in the industry such as VMware, Hyper-V or one of the Linux hypervisors. Some vendors have taken the Linux hypervisors and created their own, customized version of the hypervisor.

The network is the least radically affected of the three elements being converged. In most configurations, there is no longer a dedicated storage network because there is no longer a dedicated shared storage system. But there is a significant amount of networking involved, especially at scale -- nodes have to communicate, and storage has to be allocated. Many hyper-converged architectures end up dedicating network cards and switch ports to this traffic, essentially "unconverging" the network element. Network management functions are sometimes integrated so that configurations can be set from within the converged management interface.

The storage infrastructure is the most radically impacted by hyper-converged architectures. Typically the shared storage system is eliminated, and internal storage within the nodes is aggregated into a virtual volume. The storage part of hyper-converged architectures is a software-defined, scale-out storage system that can run as a component of the hypervisor architecture.

The storage hardware is still important in the hyper-converged architecture design, and the parts are similar to a hardware-based storage system. First, the organization needs to make sure that the shared resources (CPU and network) have enough bandwidth to drive all the storage functions as well as all the virtual machines it may handle. Second, the media that will be installed inside the servers is important, making sure there is the right balance of flash and hard disk drives so that cost and performance are in sync is critical.

While hyper-converged architecture is often delivered as a single component, there are layers beneath the onion. The more of those layers that IT professionals understand, the more they will be able to select the right product for their organizations as well as adapt the architecture to future needs. In upcoming columns I will dissect each part of the hyper-converged architecture anatomy to help planners understand the various components of the seemingly converged architecture.