Hyper-converged infrastructure has become valuable to enterprise customers, and the core HCI benefits are well-suited...
to building a cloud platform.
It is unlikely that Google or Amazon Web Services will replace their bespoke servers with HCI. The economics of running a cloud service provider (CSP) are very different from the economics of enterprise IT. The workloads that are applied are also very different.
But there are many uses for small cloud platforms built on HCI. These smaller clouds might be specialist public clouds, or they might be private clouds within large organizations. The growth capabilities and scale-out nature of HCI can provide a number of hyper-converged cloud benefits.
With traditional cloud platforms, a large amount of infrastructure is deployed before the first customer pays. Often, the CSP must sell 50% of its capacity before it can recoup the cost of building the platform and realize any hyper-converged cloud benefits. Usually, the CSP deploys the platforms as a fixed-size cluster, but in a small cloud, it is easy to get the fixed size of that cluster wrong. Platform sizing is a risk in cloud deployment -- too large, and it is never profitable; too small and it becomes overloaded quickly.
In contrast, HCI clusters scale out by nodes, making it easy to start a hyper-converged cloud with a small cluster and expand it with the workload. When a tenant wants 200 virtual machines (VMs), a CSP may only need three HCI nodes. To accommodate 1,100 VMs, that CSP might deploy nine more nodes. With HCI, the unit of scale is one node: The cluster can easily grow over time as the demand grows. Because there are fewer unsold resources, the cloud platform can reap greater hyper-converged cloud benefits and be profitable for a much longer portion of its life.
One of the challenges for companies using the public cloud is the unpredictability of performance, often due to noisy neighbors. These are VMs that belong to one tenant that use the same physical hypervisor hosts, networks or storage arrays as another tenant's VMs. One tenant's VMs consume more of the shared resources, which leads to poor performance for the other tenant.
Using an HCI platform to support a cloud means that both compute and storage performance scale out with the HCI nodes. With many HCI implementations, the majority of the VM's storage access is inside the node, leading to fewer potential noisy neighbor conflicts. Similarly, HCI nodes tend to be dual-socket servers, limiting the number of VMs that reside on a single node.
One of the defining features of a cloud platform is user self-service, in which tenants deploy their own VMs using a web portal. Integrating the infrastructure into this portal can be a significant task for a cloud provider.
A common feature in HCI platforms is the availability of REST interfaces, which simplify the integration with a web portal, and thereby ease storage management. HCI platforms are designed to be controlled through software such as the self-service portal, so integration with an existing cloud management system should be simple for a hyper-converged cloud.
Building the hyper-converged cloud
When IT admins build a cloud, they should have an idea of the initial capacity, but with an HCI platform, they can start with just three physical servers. The IT admins can add more if plans call for more initial capacity. One of the many hyper-converged cloud benefits in this case is that the cluster scales out and additional capacity is available in the cloud management portal.