Kubevirt and the Cost of Containerizing VMs
summary
Benchmarks run on a single server with 32 physical cores, Ubuntu host and guest images (OpenSUSE Leap 15.2) compare KVM (libvirt/QEMU built from source) and KubeVirt 0.44 with QEMU 5.2 and libvirt7. Experiments use 1‑vCPU and 4‑vCPU VMs, 8 GiB RAM, mmtests/NAS/STREAM/iozone and stress workloads, toggling vCPU pinning, virtual topology, hugepages and native IO. Perfect guest→host topology mapping consistently gives best throughput and latency; KubeVirt suffers large losses when Kubernetes CPU manager creates mismatched pinning; under high host load differences largely disappear.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:54Overview
- 1:21What is Kubevirt
- 2:35Experimental setup
- 6:00Improved performance
- 12:34Benchmarks
- 15:41Firm Bench
- 17:46Conclusion
related talks
31mKubeVirt: What If Kubernetes Orchestrated Your VMs?
20mKubeVirt Benchmar...
43mKubeVirt - Kubernetes, Virtualization and Your Future Data Center
32mA journey through supporting VMs with dedicated CPUs on Kubernetes
43mMaking Running Virtual Machines in a Kubernetes Cluster a Mainstream Activity
25mKubeVirt: Converge IT infrastructure into one single Kubernetes platform
32mKubeVirt - making running virtual machines in a k8s cluster
31mVMs + Containers = The Perfect Wedding
