0 views collected around this technical thread.
This article investigates why moving applications from virtual machines to containers on Kubernetes can degrade performance, presents benchmark comparisons, analyzes the impact of network architecture and soft interrupts, and proposes optimization strategies such as ipvlan and Cilium to restore efficiency.
The article examines why applications experience slower performance after being containerized on Kubernetes, presenting benchmark comparisons, analyzing increased soft‑interrupt overhead due to the Calico ipip overlay, and proposing network optimizations such as ipvlan modes and Cilium to restore efficiency.
This article details the design, implementation, and production rollout of a container VPC solution that leverages ipvlan for CNI, a custom IPAM integration with OpenStack/Ultron, and extensive operational fixes to achieve reliable, high‑performance networking for Kubernetes workloads across multiple clusters.
This article describes how 360 unified Kubernetes pod networking with OpenStack VPC by developing a custom CNI plugin that leverages Neutron elastic ports, IPVLAN L2 mode, and OVS to achieve layer‑2 connectivity between VMs and pods, including detailed implementation steps and command examples.
The article investigates why applications experience higher latency and lower QPS after moving from virtual machines to Kubernetes containers, analyzes the root causes such as increased soft‑interrupts caused by the networking stack, and proposes optimizations like ipvlan, macvlan, and Cilium to restore performance.
This article examines Kubernetes networking by defining underlay and overlay models, describing how underlay networks like flannel host‑gw and Calico BGP operate, explaining IPVLAN/MACVLAN virtualization, and reviewing common tunnel protocols such as VxLAN and IPIP used by various CNI plugins.
This study evaluates Kubernetes container networking options—Docker host networking, Flannel, IPvlan, VXLAN, host‑gw, and AWS‑VPC—by measuring latency, throughput, and installation cost under high‑load conditions, revealing IPvlan as the lowest‑latency, highest‑throughput solution while host‑gw offers a balanced, low‑dependency alternative.