Hyperconverged infrastructure (HCI) is a way of building data center infrastructure in which each server node in a cluster provides both computing power and storage, and software pools those resources across the cluster. In the traditional, VM-oriented model a hypervisor runs virtual machines on each node, and a software-defined storage layer turns the nodes’ local drives into shared storage; some platforms also support container-focused or other models. The exact architecture varies by product, but the cluster is typically managed from one console and grows by adding nodes.
At a glance
- HCI combines compute, storage and virtualization on standard servers, usually removing the separate storage array of a traditional three-tier design.
- Capacity usually grows by adding nodes, which add compute and storage together.
- Data is protected by copying it across nodes, so the cluster is designed to keep running through a node failure.
- It is sold as software on certified hardware, as integrated appliances, or as a managed or hosted service.
What problem it solves
The traditional way to run virtualized workloads uses three separate tiers: servers for compute, a storage area network (SAN) or other shared storage array, and the network that connects them. Each tier has its own vendor, management tools, support contract and refresh cycle. Sizing the storage array is a bet on growth several years out, and specialist skills are needed to keep everything compatible and tuned.
HCI collapses those tiers. Compute and storage sit on the same servers and are managed together, so a small team can run the environment without separate storage expertise. Instead of buying a large array up front, you buy a cluster sized for current needs and add nodes as you grow. Failures and upgrades are handled node by node, often without downtime for the workloads.
How it works
Nodes. Each node is a server with processors, memory, local drives (usually flash, sometimes with hard drives for capacity) and network interfaces. Hardware options and architecture vary by product. Clusters commonly start at a small number of nodes, with the minimum set by the platform.
Hypervisor or workload layer. In traditional VM-oriented HCI, each node runs a hypervisor that hosts virtual machines. Some platforms support a single hypervisor, others several, and some also support container-focused or other workload models.
Software-defined storage. A storage layer pools the nodes’ local drives into shared storage that the cluster’s workloads can use. It writes copies or parity data across nodes so that the loss of a drive or a node does not lose data, within the limits of the protection setting. Many platforms add deduplication, compression and snapshots.
Networking. Nodes connect over a fast Ethernet network, which carries both workload traffic and storage replication between nodes. Network design matters more in HCI because storage traffic depends on it.
Management. A single console handles provisioning, monitoring, upgrades and capacity planning across the cluster. Spare capacity is usually reserved so the cluster can absorb a node failure, a form of N+1 redundancy.
HCI can run in your own facility, in colocation, at branch sites, or be delivered as a hosted service. For help comparing platforms and delivery models, see our Hyper Converged Infrastructure solution page.
When it matters for buyers
- Storage array or server refresh. When several tiers reach end of life together, it is a natural time to evaluate HCI.
- Virtualization licensing changes. Changes to hypervisor licensing or bundling can prompt a fresh look at the whole platform, including HCI options built on other hypervisors.
- Small IT teams. One system to patch and monitor reduces the specialist knowledge needed.
- Branch and edge sites. Small clusters can run local workloads where a full three-tier stack would be excessive.
- Cloud repatriation or private cloud. HCI is a common foundation when workloads move back from public cloud or onto a private cloud.
Questions to ask vendors
- What is the minimum cluster size, and how does licensing scale per node, per core or per terabyte?
- Which hypervisors and workload types (virtual machines, containers) does the platform support, and what does switching cost?
- How many simultaneous node or drive failures can the cluster tolerate at our chosen protection setting, and how much usable capacity does that leave?
- Can we add storage-heavy or compute-heavy nodes, or attach external storage, if growth is uneven?
- What network speed and switches do you require between nodes?
- How are upgrades handled, and do they require workload downtime?
- What backup and disaster recovery integrations are supported?
- What is the hardware support model, and who do we call for a problem that spans hardware and software?
How it differs from a storage area network (SAN)
A SAN is a separate, dedicated storage system that many servers reach over a storage network; the storage and its management are independent of the servers. HCI has no separate array: storage lives inside the servers and is pooled by software. A SAN lets you scale storage and compute independently and suits some very large or specialized workloads, while HCI trades that flexibility for simpler management and incremental growth. Some environments use both.