| Optimizing live-migration for minimizing packet loss with KubeVirt and kube-ovn, by Zhuanlan & Zhangbingbing, (China Mobile(SuZhou)Software Technology Co.,Ltd) | We expose more detailed steps for virtual machine live-migration in KubeVirt. Which enable kube-ovn use the multi_chassis_bindings feature as will as triggering ACL policy changes based on RARP. In this way, we significantly reduces packet loss during live migrations. |
| KubeVirt Observability with Prometheus and Grafana, by Ananda Dwi Rahmawati (Activate Interactive Pte Ltd) | As KubeVirt adoption grows, so does the need for effective observability. This talk delves into the essential metrics and tools required to ensure the health and performance of your KubeVirt environments.
We'll explore practical strategies for monitoring KubeVirt workloads, identifying potential issues, and gaining valuable insights into your virtualized environment. |
| Want a VM with local storage? Fine, just try it with Hwameistor! By Pang Wei & Zhou MingMing (QiAnXin) | There are many benefits to use VMs with local storage, like:
- Superior performance with local disk IOPS/latency/throughput.
- Less complexity compared to network-based storage system.
- Extremely low cost, VMs can be created easily even in an all-in-one k8s cluster.
I'm glad to introduce a good way to do this. With the help of [HwameiStor](https://github.com/hwameistor/hwameistor), VMs can be smoothly used with local storage(LVs or raw disk). |
| Real-Time Network Traffic Monitoring for KubeVirt VMs Using OVN and Switchdev SR-IOV VFIO Interfaces, by Girish Moodalbail & Vengupal Iyer (Nvidia) | When the data plane is offloaded from the Linux kernel to Smart NICs that support the Linux SwitchDev driver model, the ability to perform real-time monitoring of network traffic for debugging or anomaly detection is lost. This issue is exacerbated when using legacy non-switchdev SR-IOV Virtual Functions (VFs), where packets sent through these VFs are only visible on the next hop switch. Consequently, any debugging efforts would require collaboration with network administrators managing the switches. Additionally, performing real-time monitoring of network traffic for anomaly detection on switches becomes significantly more challenging.
In this presentation, we will explore how to achieve real-time monitoring of KubeVirt VMs/VMIs that are multi-homed with Switchdev SR-IOV VFIO interfaces using the open-source Open Virtual Network (OVN) SDN based on Open vSwitch (OVS). The API is defined by OVN-Kubernetes CNI as a Kubernetes Custom Resource. We will introduce the OVN packet mirroring feature, which enables the capture of packets from these interfaces and their forwarding to Scalable Functions (SFs) on the host. This process can be performed at wire speed thanks to NIC accelerators, allowing for the execution of tools like tcpdump, sFlow, and deep learning inference on the captured packets. |
| Enhancing KubeVirt Management: A Comprehensive Update on KubeVirt-Manager, by Marcelo Feitoza Parisi (Google Cloud) | KubeVirt-Manager continues to evolve as a pivotal tool for simplifying and streamlining the management of virtualized environments with KubeVirt. In this session, we'll present a comprehensive update highlighting the latest enhancements and features designed to further empower administrators and users, and help accelerate even more the adoption of KubeVirt. From improved visibility with unscheduled Virtual Machines grouped in the Virtual Machines page, to a revamped NoVNC implementation supporting complete console functionality, our latest release promises a more seamless experience.
Additionally, we introduce a new detailed information page for Virtual Machines and Virtual Machine Pools, providing insights into crucial details such as operating system version, kernel version, networking configuration, and disk specifications. Not stopping there, we also present enhancements extend to the Virtual Machine creation process, now offering options to select cache mode and access mode for disks, as well as the ability to utilize multiple network interfaces. Virtual Machine Pools receive a significant upgrade with support for Liveness Probes and Readiness Probes, alongside the introduction of auto-scaling based on Kubernetes HPA. Moreover, our integration with Cluster API provides now a streamlined option for administrators and users to bring new Kubernetes clusters with just a few clicks.
Join us as we delve into these exciting updates to facilitate the management of KubeVirt environments. |
| Hack your own network connectivity, by Edward Haas & Leonardo Milleri (Red Hat) | KubeVirt provides common network connectivity options for Virtual Machines that satisfy most scenarios and users. As the project popularity grows, special scenarios and needs surface with requests to tweak and customize network details. The sig-network was faced with a double challenge, to satisfy the growing community needs and at the same time to keep the codebase in a maintainable state.
The [network binding plugin](https://github.com/kubevirt/kubevirt/blob/main/docs/network/network-binding-plugin.md) has been introduced to solve these challenges, giving the ability to extend the network connectivity in KubeVirt and at the same time assist in maintaining the core network functionality.
We will present the pluggable infrastructure and demonstrate a success story that used it to extend the supported interfaces (vDPA). |
| Introducing Application Aware Resource Quota, by Barak Mordehai (Red Hat) | If you want to hear about how Application Aware Resource Quota solves KubeVirt's Quota related issues during migrations and obscures Virtual Machines overhead from Quota - this session is for you.
You will also hear how this project opens the door for plug-able policies to empower other operators to customize resource counting. |
| Kubevirt enablement on IBM Z and LinuxONE, by Nourhane Bziouech (IBM) | Kubevirt keeps evolving and growing in features and capabilities , and one of the growth aspects is the platforms where it runs. This session will cover the usage of Kubevirt for s390x architecture when running it on IBM Z and LinuxONE.
Join me in this session and let me take you through the journey of adding Kubevirt to the IBM Z and LinuxONE platform. |
| Ensuring High Availability for KubeVirt Virtual Machines, by Javier Cano Cano (Red Hat) | Maintaining high availability of Pod workloads within Kubernetes is a common objective, typically achieved using Kubernetes' built-in primitives. However, certain workloads, such as KubeVirt Virtual Machines (VMs), necessitate at-most-one semantics, which can only be achieved with the assistance of additional operators like Node Health Checks (NHC) and specific remediation operators.
This presentation will explore the methods for establishing a high-availability infrastructure specifically designed for KubeVirt virtual machines (VMs). The discussion will encompass the development of a testing methodology for KubeVirt in conjunction with the Node Health Check (NHC), including the assessment of VM recovery times in the event of node failure. The presentation will detail the challenges encountered, the debugging process, and the solutions implemented, with a focus on the current status of VM recovery times. |
| Coordinated Upgrade for GPU Workloads, by Natalie Bandel & Ryan Hallisey (Nvidia) | In this talk, we will explore how the Nvidia GeForce Now platform manages maintenance and upgrade activities on Kubernetes clusters with KubeVirt workloads by leveraging native Kubernetes components.
Ongoing maintenance is essential for any Kubernetes cluster. The complexity of coordination and management of the maintenance activities only increases for Kubernetes clusters with KubeVirt and thousands of VMIs running simultaneously. In this talk, we will explore how the Nvidia GeForce Now platform leverages native Kubernetes components to manage and coordinate maintenance activities on Kubernetes clusters with thousands of GPU workloads across multiple data centers. We will present the architecture and benefits of our approach, explaining how we maintain a single source of truth for real-time status updates from the Kubernetes cluster. We will discuss our efficient scheduling algorithm, which takes into consideration existing VMI workloads for prioritizing maintenance tasks. We will cover our validation and failure handling processes. And finally, we will highlight actual improvements of operational efficiency in maintenance times during GeForce Now upgrades. |
