Building your own IT infrastructure is a complex and often thankless task, especially when your apps and users expect to work in a cloud-like environment with fast deployment, seamless reliability and ready scalability. You need to buy servers, storage and networking gear; install virtualisation, provisioning and orchestration...
software; and then administer and update it all as it grows and evolves.
Hyper-converged infrastructure is intended to offload much of that work. All those layers and components are abstracted into a single system, on to which you install your virtual machines (VMs) and applications.
Server clustering and storage pooling are typically built-in too, and allow you to scale up, or more commonly out. Many systems will also let you add and re-use existing servers and storage as well.
The aim is to do what the webscale cloud providers do, but instead of having to build it from scratch, hyper-convergence puts an enterprise cloud in a box as a turnkey appliance.
It drives a conceptual change as well. Instead of focusing on virtualisation or containers, as IT architects and suppliers have done it the past, hyper-convergence is application-centric.
This is a significant change, and it has helped open the door for those eager to win a slice of this new market, as has the fact that hyper-convergence is largely a matter of software.
More specifically, it is very much software-defined, which allows you to swap or expand hardware layers more easily and cheaply than you can with discrete or proprietary hardware. It also permits some hyper-converged systems to support mixed clusters, for maximum expansion flexibility.
So although the companies listed here can all provide a turnkey appliance, they are all really software companies, not hardware developers. They typically work with hardware providers – some with one, others with several – who build off-the-shelf boxes on to which they load their software.
A few even offer free but unsupported (or community-supported) versions of their software for you to install onto your own hardware. The key word here though is unsupported. The main reason they prefer to sell complete appliances built on known and standard hardware platforms is to minimise their support workload.
Read more about hyper-converged infrastructure
With its flagship USX software-defined storage platform now maturing well, Atlantis Computing comes at hyper-convergence from the perspective of storage. It is highly scalable and flexible and can create hyper-converged infrastructure from existing resources by pooling server-side flash and bringing existing SAN and NAS capacity into its distributed file system, for example.
Based on USX, HyperScale is Atlantis’s clustered, flash-only, hyper-converged appliance product. Each CX appliance contains four server nodes, giving it up to 96 cores and 3TB of RAM, plus 12TB or 24TB of all-flash storage. The cluster can run Citrix XenServer or VMware vSphere virtualisation environments.
Atlantis does not build its own appliances. Instead, channel partners go to one of the hardware providers Atlantis works with, namely Cisco, Dell, HPE, Lenovo and SuperMicro, then assemble and deliver a complete system based on Atlantis-specified components and bundled with a full hardware and software support package.
Maxta’s MaxDeploy is an appliance based on the company's MxSP software. It majors on flexibility and lack of lock-in. The company says it can support any hypervisor and storage, with VM-centric storage management and the ability to also bring existing servers into a hyper-converged infrastructure.
Like Atlantis, Maxta works with channel partners and with Cisco, Dell, HPE, Lenovo, Quanta and SuperMicro as hardware suppliers. A configuration tool on the Maxta website enables you to specify your requirements, which are then combined with Maxta’s reference architectures to generate either a pre-configured or a custom-built appliance using your chosen brand of hardware.
MxSP aggregates and pools server storage under a VM-centric global namespace. It allows you to scale up or out, and to scale the storage and compute pools independently. A higher-end MaxDeploy system might feature six two-socket servers, for up to 144 cores and 768GB of memory, plus 12TB of mixed hard disk and SSD, but clusters can also scale up to 24 servers.
Nutanix Xtreme Computing Platform
Nutanix was the first hyper-converged rival to get under VMware’s skin and is notable for goading the virtualisation giant into a battle of the blogs in 2015.
Its Xtreme Computing Platform (XCP) is a family of scale-out appliances with Nutanix's own Acropolis infrastructure software and distributed storage fabric. Each appliance can contain up to four server nodes and can add up to 38TB of SSD and 60TB of hard disk to the cluster’s storage pool.
Nutanix also makes its software available for OEMs such as Dell to integrate, and offers mobility tools to help customers convert applications from VMware and Hyper-V to Acropolis KVM, which it says could significantly reduce licensing costs – the vTax, as it is often called.
Lastly, Nutanix has developed its own automation and orchestration toolkit called Prism, which greatly simplifies the deployment of applications and VMs. This is an area where KVM has been relatively weak. It offers a free software-only community edition of Acropolis for users to evaluate on their own hardware.
Scale Computing HC3
Notable for its high degree of integration and abstraction, and its consequent ease of use, Scale’s HC3 is a turnkey datacentre-in-a-box for small to medium-sized businesses. VM-centric software-defined storage, non-disruptive expansion, and automated high-availability, data tiering and workload placement, all contribute to simplicity of administration.
The HC3’s HyperCore operating system includes a centralised and unified management layer that ties together the various elements within the box and drives deployment and provisioning, monitoring, management and so on. HyperCore, which is now in its sixth generation, also includes a free KVM-based hypervisor.
You need at least three HC3 appliances for a cluster and the company says it has tested clusters of up to eight nodes.
The appliances scale from the entry-level HC1100 with six cores, 64GB of memory and SAS hard disks (an SSD in each is optional) to the top-of-the-line HC4150 with 16 cores, faster processors, 384GB of memory, up to 12TB of disk and 1.6TB of SSD.
A purpose-built hyper-converged appliance, OmniCube majors on storage optimisation and close integration with VMware. It has its own distributed file system and pooled data store, plus hardware to accelerate processes such as data compression and deduplication. As a result, the company guarantees 10:1 data optimisation and claims 40:1 is feasible, which of course translates to lower costs and greater efficiency.
A cluster will typically have three OmniCube appliances, although you can run smaller sites with just one and still get the benefits of simpler deployment and admin. Simplivity’s management tools are globally federated, allowing you to manage, automate and orchestrate all appliances, remote and local. Backup and recovery features are built in too for remote and local data protection, with OmniCube’s data efficiency permitting synchronous replication even over a kilometre or two.