While most instances of Kubernetes today are deployed on virtual machines running in the cloud or on-premises, there is a growing number of instances of Kubernetes being deployed on bare-metal servers.
The two primary reasons for opting to deploy Kubernetes on a bare- metal server over a virtual machine usually are performance and reliance on hardware accelerators. In the first instance, an application deployed at the network edge might be too latency-sensitive to tolerate the overhead created by a virtual machine. AT&T, for example, is working with Mirantis to deploy Kubernetes on bare-metal servers to drive 5G wireless networking services.
In the second case, organizations such as Bloomberg L.P. have employed graphical processor units (GPUs) and other hardware accelerators that don’t lend themselves to virtual machines.
While both these types of use cases involving Kubernetes are still comparatively nascent, Brian Gracely, director of product strategy for Red Hat, says it’s now only a matter of time before more applications that employ, for example, machine learning algorithms that require direct access to hardware, which will result in more instances of Kubernetes being deployed on bare-metal servers in the cloud and on-premises.
That shift will lead to a third era of Kubernetes characterized by hybrid IT environments where instances of Kubernetes will be running on both virtual machines and bare-metal servers in near equal measure, says Gracely. In turn, many IT organizations will need to re-evaluate their current management frameworks, which in many cases are optimized for virtual machines. In fact, the number of IT professionals who have any experience with provisioning bare-metal servers, after spending two decades or more focused on virtual machines, is relatively low.
Less clear is to what degree Kubernetes cluster might one day subsume the need for virtual machines altogether. Kubernetes unifies the management of compute, storage and networking within a cluster. Many organizations prefer to deploy it on virtual machines today for one of two primary reasons: they simply don’t have the tools to manage Kubernetes on bare-metal servers, or they value the isolation virtual machines provide between instances running on the same physical machine. To address that latter issue, several initiatives are underway to create much lighter-weight virtual machines to deploy Kubernetes in place of traditional virtual machines such as VMware or Kernel-based Virtual Machines (KVM).
VMware rivals have been especially keen to raise the issue of a “VMware Tax” to highlight the cost of having to license commercial virtual machines from VMware to run an open source stack of software. VMware counters that it provides an approach to operationalize Kubernetes within an already complex enterprise IT environment.
When it comes to choosing between virtual machines and bare-metal servers, most enterprise IT organizations will split the difference based on the characteristics of the application workloads being deployed. But the number of IT organizations that need IT professionals with knowledge of managing bare-metal servers most certainly will increase.