Hyperconverged infrastructure (HCI) is taking the data center by storm. In fact, a recent survey conducted by 451 Research found that 40 percent of IT organizations have already deployed an HCI platform.
HCI arguably represents a natural progression from converged infrastructure. The first converged infrastructure systems unified the management of compute, storage, and networking by pre-integrating them in a rack platform. HCI takes that concept a step further by seamlessly unifying the management of compute and storage using a scale-out architecture based on appliances.
As HCI appliances scale out, they also put a lot of extra pressure on the networking environment. Once HCI appliances begin to proliferate, the network becomes the backplane over which all the I/O traffic generated by these appliances travels. In fact, as the HCI platform makes use of all-flash storage technologies plugged into modern NVMe controller interfaces, the amount of data these systems can process will increase exponentially. Before too long, IT organizations discover that their legacy switches are no longer capable of keeping up with I/O requirements.
It’s hard to say with absolute certainty just what level of performance will ultimately be required. Many data centers are already moving to embrace either 10G or 25G Ethernet switches. But as data centers continue to scale out using HCI systems, it might not be long before many of them need 40G/50G or even 100G switches. Of course, because IT budgets are limited it’s likely that IT organizations will find have to simultaneously deploy 1, 10, 25, 40, and 50G switches in different classes of data centers.
Switches, however, are not the kind of IT infrastructure that is easily replaced. In fact, switches are now an architectural decision. IT organizations need confidence in their ability to upgrade network switches to meet performance requirements with the least amount of friction possible.
In the age of HCI systems, IT organizations need to view network switches as a platform decision. In the past, acquiring a switch was a one-time event that occurred every five years. Scale-out computing architectures enabled by HCI systems will accelerate the rate at which most enterprise networks will need to regularly upgrade their network switches.
Over time IT organizations are likely to embrace multiple HCI platforms. In fact, there’s already no shortage of HCI options. HCI systems from Nutanix, Dell EMC, Cisco and Hewlett-Packard Enterprise (HPE), for example, are likely to be employed at one time or another by the same IT organization. A common switch fabric is critical to allowing IT organizations to keep their HCI options open in a way that offloads as much of network processing from those HCI systems as possible. For this to happen, the network switch needs to be able to seamlessly integrate with any number of software-defined networks that will provide the fabric across which multiple HCI platforms will be connected as part of a modern scale-out architecture.
Of course, not every application workload lends itself to an HCI platform. Rack-based systems that scale compute, storage, and networking will be deployed alongside appliances for years to come. Depending on the nature of the application workloads being supported, IT organizations will find themselves needing to deploy systems that scale both up and out. The real issue will be finding the best way possible to treat all these systems equally on networks they all need to share.