Once you decide to deploy hyper-converged infrastructure as your virtualization framework, you must confirm that...
your data center's network and server racks can provide the proper support. Implementing hyper-converged infrastructure does require planning, but the prerequisites are relatively simple.
HCI vendors tout their products as being easy to implement and operate, but few HCI vendors go into much detail about the infrastructure that is required before you run the setup wizard; most of the specifications are for physical space, network connections, power cables and hypervisor software.
But there are usually some dependencies on network and hypervisor management that you must plan for before you deploy an HCI appliance. If you work with a service provider, you can ask them about adding any necessary updates to your service license agreement for installation, configuration and maintenance.
Setting up the network
Most HCI appliance deployments require 10 Gigabit Ethernet (GbE) for the storage network to deliver robust storage performance. Each HCI node needs at least two 10 GbE switch ports, which are for the HCI storage cluster, as well as for VM migrations. They also support management and VM networking.
HCI nodes also often require two 1 GbE switch ports per node to keep management and VM networking separate from the storage network. Each node also needs a single 1 GbE switch port for out-of-band management.
Each HCI node needs five network switch ports and cables. If you have the typical four nodes in a 2U HCI chassis, then there should be 20 network cables in that same 2U. There are four network traffic types you'll support: storage, migration, management and VM.
You must isolate these traffic types from one another for secure user access and optimal network performance. You can achieve isolation with virtual local area networks (VLANs) configured on the physical and virtual switches. You must set up these VLANs before starting the HCI deployment.
If you can't use VLANs, you'll need more network adapters in the HCI nodes and more network cables and switch ports. Check your HCI vendor's documentation about routing and firewall requirements between these networks if you want full connectivity; the management network often needs to reach the out-of-band network to enable full manageability.
Plan for a power increase
Hyper-converged deployments include rack servers, so make sure you have enough rack space in your data center. Often -- but not always -- HCI appliances are deployed as chassis that have four server nodes in 2U -- 4-in-2U -- which is very dense; do not assume that twelve nodes of HCI mean six units of rack space -- 3 x 4-in-2U. There are HCI appliances that use discrete 1U or 2U servers, as well as 2-in-2U enclosures. Some vendors even offer all of these options in their catalogues.
A 12-node cluster can be six to 24 rack units in height, and each server or enclosure requires two power cables for redundancy, power safety and downtime elimination. A 12-node cluster needs anywhere from six to 24 power cables, and each cable could require up to 2,400 watts of power.
Make sure your power distribution unit's racks and data center can handle this dense power load. You might be tempted to put your new HCI nodes all in the same server rack, but most HCI deployments are more fault tolerant if you spread out the nodes. Over time, you can track power loads with data center infrastructure management software and adjust the cooling units accordingly.
By being aware of rack layouts, the HCI platform can lay data across the nodes so your VMs are still available, even if all the nodes in one rack are offline. Splitting nodes evenly across at least three racks is a good idea to avoid losing the whole HCI setup if there is a power or network fault in one rack.
Managing hyper-converged infrastructure
Some HCI vendors and tutorials assume you're building infrastructure from the ground up, which is excellent if you start from nothing. Most organizations that deploy HCI appliances already have a data center infrastructure and they want to add HCI to their existing architecture. The new platform needs connectivity and software resources from the current infrastructure. This can be as simple as needing the available IP addresses on existing subnets and domain name systems to match the new HCI nodes.
Each hardware node needs at least one IP address on the management, storage and out-of-band management networks. Most HCI offerings use a third-party hypervisor for VM management, such as vSphere or Hyper-V. You'll need to license the hypervisor and integrate it with third-party management software, such as PowerShell Hyper-V Manager or HyperV_Mon. A small number of HCI products -- such as Nutanix AHV -- automatically include hypervisor management functionality.