Hyper-converged infrastructure is a hot topic across several IT disciplines. A typical hyper-converged infrastructure appliance brings compute, storage, hypervisors and occasionally networking together into a single modular system. All of this capability falls under a common management console through a single pane of glass.
Often referred to as an infrastructure in a box, a hyper-converged infrastructure appliance combines all the pieces necessary for data center computing. Some hyper-converged device vendors even assert that their appliances offer a more advantageous total cost of ownership, with lower Opex and Capex costs than cloud-based technologies such as Amazon Web Services. Most of those vendors rightly tout the reduced costs of administration, thanks to the all-embracing console with its unifying setup, configuration, management and monitoring capabilities.
Hyper-converged appliances scale linearly, which makes it easy to add nodes. And because IT doesn't have to configure servers, storage, hypervisors and network devices separately, the initial hyper-converged infrastructure appliance can be implemented easily. However, there is a drawback when it comes to flexibility -- a customer who needs one resource needs to add others as well. For example, if data storage capacity runs low, a customer must pay for a new appliance that also includes compute and hypervisor licenses, even if they are not required.
Some hyper-converged device vendors -- such as Gridstore -- offer graphics nodes that can make high-performance, drop-in graphics engines available for high-end graphical processing, usually for virtual desktop infrastructure workstations or rendering client use.
What's in a hyper-converged infrastructure appliance?
Key components in hyper-converged appliances include:
- CPUs to run the native hypervisor, automation, and management tools and utilities, rafts of virtual machines (VMs), as well as virtualized storage and networking functions. From four to 16 multicore CPUs are typical per appliance.
- RAM to provide work space that keeps all the moving software parts running quickly and efficiently. Typical configurations contain from 64 GB to 512 GB of RAM.
- Storage to provide room to store VMs, applications and associated data.
Analyst George Crump discusses the benefits and drawbacks of hyper-converged configurations such as software-only and flash.
These resources are set inside blades or rack-mount systems, usually with dual hot-swappable power supplies and hot-swappable storage devices. Storage is usually a mix of flash solid-state drives and conventional spinning disks, though the tendency is to move to as much solid-state storage as customers' pocketbooks will permit. Several hyper-converged vendors sell all-flash products.
Software is key
Software is the key to the allure and value of hyper-converged systems. Modern virtualization technology goes a long way toward explaining how and why hyper-convergence works, adds value, and makes a compelling consideration for data center and server room use. Most hyper-converged appliances support VMware vSphere or ESXi hypervisors, but open source KVM and Microsoft Hyper-V versions are available. Customers must purchase software licenses from VMware or Microsoft when using their hypervisors.
Hyper-converged software usually builds in data protection schemes. This can include mirroring, striping, replication and erasure coding to add reliability and to help keep overall data volumes down -- though they're still considerable, given that these systems provide access to massive storage. Other interesting aspects of the software include automation support, the ability to perform rolling updates without taking VMs offline, VM migration and management tools, and many varieties of load- and resource-balancing techniques. Most hyper-converged products include automatic failover support so that a lost compute node or storage device won't bring down a running VM. Finally, most products include various options for backup and recovery, and disaster recovery or business continuity capabilities.
The building block model for scaling and growth
To support failover and fault tolerance, most device clusters of this kind require a minimum purchase of three nodes. Additional nodes can be added to the cluster at any time. Scaling for these environments varies from vendor to vendor. Some max out at 12 or 16 total nodes, while others support hundreds or thousands of nodes.
Buying hyper-converged: What and how much?
Pricing for hyper-converged appliances varies widely. For as little as $25,000, an organization can buy a modest hyper-converged data center setup, with 12 multicore CPUs, 18 TB of storage, 10 Gigabit Ethernet network connectivity, and all the software needed to stand up and run more than 900 VMs. At the other end of the spectrum, the largest configurations are capable of running hundreds of thousands -- or even millions -- of VMs, and can cost hundreds of thousands of dollars.
Some hyper-converged vendors sell software only and rely on hardware from vendors such as Dell, Hewlett Packard Enterprise and Lenovo for memory, CPU, storage and other circuitry. Some vendors support the Open Compute initiative that uses widely available commodity rack-mount servers and components, while others define reference architectures for those willing to act as "do-it-yourself" system integrators. But the biggest appeal for hyper-converged appliances, especially at the low to middle range in the market, is for turnkey devices vendors claim can be made fully operational in a matter of minutes.
Why more enterprises are switching from SAN to hyper-converged storage
Discover why hyper-converged devices challenge the primacy of SAN storage
Read our essential guide to hyper-converged storage options
How to choose the right HCI storage vendor