IBM this week expanded the scope of its ambitions for containers in the cloud by making available an instance of its managed cloud service that includes a bare-metal option for deploying Kubernetes.
Jason McGee, vice president and CTO for IBM Cloud, says that as organizations become more comfortable with containers and the clusters they run on, IBM is seeing a marked increase in the number of organizations that want to eliminate the need for reliance of legacy hypervisor forms of virtualization. Most cloud service providers and internal organizations initially deploy containers on top of hypervisors because all their tooling assumes virtual machines are the base layer being employed to abstract the underlying infrastructure.
But organizations that want to achieve better performance and higher levels of density per physical server are moving away from hypervisors, says McGee.
To further spur that transition, McGee says IBM is also working toward providing applications built on top of Kubernetes with direct access to graphical processor units (GPUs), which lie at the heart of many big data and AI applications.
In addition to a lack of tooling, concerns about the ability to fully isolate containers from nosy or noisy neighbors has pushed many organizations to deploy containers on top of hypervisors in production environments. But McGee notes that shift plays to an IBM strength because the company has been making bare-metal servers available as an alternative to virtual machines for quite some time.
The economics of deploying containers on top of hypervisors can be problematic. The number of containers that can efficiently run on top of hypervisors sharing the same physical server is a fraction of the number of containers that can run natively on a physical server. At a time when many organizations are trying to deploy more applications than ever, hypervisors can wind up being a much more expensive approach to deploying containerized applications.
McGee says frameworks such as Istio are maturing rapidly to the point where the tools for managing containers running on a Kubernetes cluster has significantly advanced, which removes one more reason why organizations should rely on hypervisors to run containers.
It’s unclear the degree to which organizations will be willing to abandon hypervisors for deploying containers in a production environment. Most cloud service providers other than IBM require Kubernetes to be invoked as a service running on top of a hypervisor. But many internal IT organizations already make use of containers on bare-metal servers when building applications. It’s only when IT operations teams are asked to deploy those containerized applications in a production environment that hypervisors get added to the equation. As containerized applications running on bare-metal servers continue to mature, it’s not likely they will eliminate the need for hypervisors altogether. But they will either reduce reliance on hypervisors or force a fundamental change to the way hypervisors are constructed.
In the meantime, the debate concerning the continued relevance of hypervisors in the age of containers is likely to remain heated for quite some time. IT operations teams have invested heavily in virtual machine software for decades. That said, commercial instances of hypervisors such as VMware can be expensive, which in part may also be a big reason that there’s so much newfound interest in bare-metal servers.