The hyper-converged network becoming more common in HCI
It is commonly said of hyper-converged infrastructure systems that they virtualize storage, compute and networking resources. That last one, networking, is not the same as the other two in an HCI implementation. In an HCI system, it is easy to take physical compute and storage resources and turn them into logical pools of virtual resources to be delivered to the virtual machines (VMs) at the end-user level. In each of those two cases, you are virtualizing just a few things — CPU cycles and RAM for compute or data memory space for storage. In contrast, trying to implement a hyper-converged network means virtualizing the physical switches and controllers and dividing up among the VMs the bandwidth, throughput, access and other features they provide. Virtualizing how the end user accesses network resources, however, is not the same thing as making the resource itself virtual. Because the network is vital to all the other virtualized resources, it is arguably the most important aspect of an HCI system. If the network function doesn’t operate properly, the VMs can experience everything from slowdowns to simply not being available to the end user. In other words, no network means no VMs to work on. One way a hyper-converged network is used by nearly all HCI platforms is by virtualizing the network interface card (NIC) in the VMs. This is also known as I/O virtualization, because it virtualizes the network input and output of the VM. Both @Microsoft’s #HyperV hypervisor and VMware’s ESX support this, and Microsoft added NIC teaming to Windows Server 2016, allowing virtual NICs to be combined into a pooled virtual network adapter to reduce connectivity and throughput issues for any one VM. Virtualizing how the end user accesses network resources, however, is not the same thing as making the resource itself virtual. Virtualizing networks To be truly on the same level as the compute and storage resources in a hyper-converged platform, your hyper-converged network would need to take the function of the physical switches and controllers. It would have to turn them into a logic pool from which any set amount could be assigned to each VM. That would require much of the network hardware to be contained in the HCI appliance or replicated via software in the appliance. VMware is one of the vendors with the most fully formed hyper-converged network technology, via its NSX product. While NSX was originally developed to work solely in a VMware HCI system, last year, the company released NSX-T, which can bring network virtualization to operations running the Kernel-based Virtual Machine hypervisor or ones based on OpenStack. Nutanix is working on adding networking to its HCI platform through its Acropolis infrastructure stack. These networking features include microsegmentation services and orchestration through open APIs. Software-defined networking and HCI Software-defined networking (SDN) is exactly as stated: running software that virtualizes management of typical network hardware, such as switches, firewalls and routers. SDN is often based on the OpenFlow protocol, although other protocols, such as OpenDaylight and even OpenStack, can be used. One of the advantages of SDN is that it allows a network admin to manage the flow of data through a network from a central control console, instead of having to adjust the individual hardware components separately.