VMware’s Kubernetes Evolution: Quashing Complexity
Kubernetes adoption remains a major hurdle for many engineering organizations, even more than a decade after the container orchestrator was introduced.
To illustrate: While 84% of organizations say they are evaluating or deploying Kubernetes in production, 46% lack proper training and find complexity to be their biggest challenge, according to a 2023 report by the Cloud Native Computing Foundation.
“The problem is that to make Kubernetes usable, you actually need to go beyond just a Kubernetes runtime. You actually need to have a Kubernetes platform — a set of services that just are running,” said Paul Turner, vice president of products for VMware’s Cloud Foundation division at Broadcom, in this episode of The New Stack Makers.
“The problem is, otherwise, that the developer — one of the most expensive resources — has to spend a lot of their time running infrastructure services, which should just be there.”
Turner and episode host Alex Williams, TNS founder and publisher, explored how VMware is tackling Kubernetes complexity while enabling containerized workloads alongside virtual machines.
VMware’s solution is their VCF (VMware Cloud Foundation) platform, which Turner called “the new vSphere.” VCF delivers a complete Kubernetes environment with pre-configured components, including controller runtime, Harbor registry, Valero backup, networking, storage, and Istio — all configured out of the box and managed by VMware.
But doesn’t adding an abstraction layer increase complexity? “People seem to have this funny belief that virtualization has a significant overhead. It does not,” Turner said.
He backed up his statement by citing performance metrics: VMware recently released ML Perf benchmarks showing virtualized container-based AI applications running at 98.3% of bare metal performance while providing benefits like dynamic application movement across clusters.
Virtualization, Containers and AI
The conversation turned to how VMware’s approach addresses the intersection of virtualization, containers and AI workloads. According to Turner, “AI is probably driving almost half the deployments of Kubernetes out there — AI-ready applications.” This convergence has led to VMware’s Private AI Foundation joint platform offering with Nvidia, which simplifies GPU virtualization for AI workloads.
Turner spoke proudly of VMware’s GPU virtualization capabilities: “We can do completely non-disruptive V motion of AI applications running PyTorch frontends. We can move them across servers without any disruption. Everything works, and that’s with a 70 billion parameter Llama model running an inference service at the same time.”
He emphasized VMware’s continued open source commitment, noting it’s been a top-three contributor to Kubernetes while leading or founding projects like Harbor, Antrea, and Envoy, and being active in Cluster API, Contour, and Valero.
“Open source has always been close to VMware,” Turner said. “We want the Kubernetes capability to remain open standards-based, available across any ecosystem. We want it to be that application layer that your customers and any customer can rely on as a cloud-independent layer.”
Watch the full conversation to learn more about VMware’s vision for simplifying Kubernetes management, its commitment to open source, and how it’s addressing the challenges of GPU utilization in private AI environments.
