This document discusses deploying virtual network functions (VNFs) using Kubernetes pods and VMs. It covers using single root I/O virtualization (SR-IOV) and Open vSwitch with Data Plane Development Kit (OVS-DPDK) for high performance networking. SR-IOV allows VNFs direct access to network interface cards to bypass the hypervisor. OVS-DPDK processes packets in userspace using DPDK for accelerated performance compared to native Linux networking or SR-IOV for some workloads. The document provides configuration details for enabling SR-IOV and OVS-DPDK on the host and specifying network interfaces in KubeVirt virtual machine instances.

1. Deploying VNFs with
Kubernetes pods and VMs

2. Agenda
VNF basics
- What are VNFs?
- Benefits of VNFs
- Enhancing app
performance
SR-IOV apps
- What is SR-IOV?
- Host config for
SR-IOV
- VM deployment
using KubeVirt
OVS-DPDK apps
- What is OVS?
- What is DPDK?
- Host config for
OVS-DPDK
- VM deployment
using KubeVirt
Demo

3. Virtual Network Functions

4. Network Function Virtualization
NFV is a network architecture concept to abstract network functions
from hardware.
Architecture:
● Virtualized network functions (VNFs)
● Network functions virtualization infrastructure (NFVi)
● Management, automation and network orchestration (MANO)

5. What are VNFs?
● Virtualized network services that replace legacy network
appliances on proprietary hardware
● VNFs are built on top of NFV infrastructure serving as a
foundational technology for 5G or edge networks
● Often deployed as virtual machines (VMs) by various
telecommunications providers
● Common VNF applications - routers, firewalls, WAN
optimization, NAT, load balancers

6. Benefits of VNFs
● Improved network scalability
● Efficient use of network infrastructure
● Reduced power consumption
● Better security features
● Saves on physical space needed for hardware
● Reduced operational and capital expenditures

7. Enhancing VNF performance
Heavy data traffic when running multiple VNF VMs on a host.
Efficient memory access, task and resource allocations, network I/O.
Faster packet processing than native Linux kernel network stack
● SR-IOV
● DPDK

8. SR-IOV

9. What is SR-IOV?
Single Root I/O Virtualization allows the isolation of PCI Express
resources for manageability and performance reasons.
It allows VNFs to access NIC directly, bypassing hypervisor.
Requires support in BIOS and at OS level.
● Physical functions (PFs) - full-featured PCIe functions
● Virtual functions (VFs) - “lightweight” PCIe functions

11. KubeVirt support
● SR-IOV device plugin
● SR-IOV CNI plugin
● Multus meta-plugin
Ref:
https://github.com/kubevirt/kubevirt/blob/main/docs/sriov.md
https://kubevirt.io/user-guide/virtual_machines/interfaces_and_networks

12. SR-IOV host config
● Plug in SR-IOV capable NIC
● Enable SR-IOV in BIOS
● Configure kernel to enable IOMMU:
○ intel_iommu=on
○ pci=realloc
○ pci=assign-busses
● VFIO userspace driver to pass through PCI devices into qemu:
○ modprobe vfio-pci

13. KubeVirt VMI spec
spec:
domain:
interfaces:
- masquerade: {}
name: default
- name: sriov-net
sriov: {}
KubeVirt relies on VFIO userspace driver to pass PCI devices into VMI guest.
networks:
- name: default
pod: {}
- name: sriov-net
multus:
networkName: sriov-network-eno

15. OVS-DPDK

16. What is OVS?
Open vSwitch: a production quality, multi-layer virtual switch
Main components:
● Forwarding path: implemented in kernel space for high
performance
● Vswitchd: main userspace program

17. What is DPDK?
● DPDK stands for Data Plane Development Kit
● Packet processing bypasses Linux kernel network stack
● Fast switching in user space using Poll mode drivers
● Open vSwitch can be combined with DPDK for accelerated
performance
● For East-West traffic in same server, DPDK wins against SR-IOV

19. KubeVirt support
● Userspace CNI plugin
● Multus meta-plugin
● OVS built with DPDK support
Pending Github PR - https://github.com/kubevirt/kubevirt/pull/3208
Ref:
https://github.com/intel/userspace-cni-network-plugin
https://telcocloudbridge.com/blog/dpdk-vs-sr-iov-for-nfv-why-a-wrong-deci
sion-can-impact-performance/

20. OVS-DPDK host config
● Install DPDK and OVS packages on host(s)
● Configure hugepages using sysctl: vm.nr_hugepages
● Setup DPDK devices using VFIO-PCI:
○ driverctl set-override <pci-address> vfio-pci
● Bridge/Ports creation in OVS:
○ ovs-vsctl add-br br-dpdk0 -- set bridge br-dpdk0
datapath_type=netdev
○ ovs-vsctl add-port br-dpdk0 eno1 -- set Interface eno1
type=dpdk options:dpdk-devargs=0000:19:00.1

21. KubeVirt VMI spec
spec:
domain:
interfaces:
- masquerade: {}
name: default
- name: vhost-user-net-1
vhostuser: {}
KubeVirt relies on VFIO userspace driver to pass PCI devices into VMI guest.
networks:
- name: default
pod: {}
- name: vhost-user-net-1
multus:
networkName: net1

23. Demo…

24. Thank You!!