With the rising costs of hypervisor licensing, many organizations are evaluating their virtualization needs and exploring industry and market trends.
In this blog weโll explore the strategic advantages of running virtual machines on Kubernetes using KubeVirt. KubeVirt consists of a set of custom controllers that allows management of KVM virtual machines inside Kubernetes Pods leading to a unified management paradigm for any workload running on any infrastructure.
The broad stroke of why
In the 2024 Gartner Magic Quadrant for Container Managementยน report the assumption is that virtual machine management is an optional component for your container management platform. Yet, the same report distills these predictions:
- By 2027, more than half of all container management deployments will involve serverless container services, up from less than 25% today.
- By 2027, more than 75% of all AI deployments will use container technology as the underlying compute environment, up from less than 50% today.
- By 2028, 80% of custom software running at the physical edge will be deployed in containers, up from 10% in 2023.
Let's break this down.
We know that micro VMs are far more secure for serverless workloads. Boot times for VM based containers are marching towards double-digit milliseconds using optimized runtimes such as Firecracker. Hyperscalers do this today already, so the technology is ready for enterprise private clouds.
Ideal for single tenant deployments, managing GPU resources for AI workloads in container runtimes is nothing new and has been available for many mainstream AI applications such as Hugging Face. For isolation and autonomy of platform engineering teams, the GPU resources can be securely compartmentalized with VMs using the GPU Operator for KubeVirt.
Edge workloads has benefited from containerized workloads for over a decade now through easy ubiquitous management of workloads across fleets of outfits such as retail, surveillance and manufacturing. VMs at the edge allow practitioners to utilize another abstraction layer while keeping the same control plane for both VMs and containers using KubeVirt, for either Intel and ARM-based deployments.
The bottom line is that the contrast between the containerized application and the VM workload that weโve kept around does not need to be on a split horizon. Letโs understand the patterns on how we got here and how we need to think about the future in terms of strategy and partnerships.
Application modernization
The decade old imperative to modernize applications to either move to the public cloud, use twelve factor patterns or refactor for container runtimes has become the new standard. Yet, virtual machines on private and hybrid clouds are sometimes more important than ever to manage the most trivial workload.
The cornerstone net upsides from combining the two paradigms are numerous:
- Unified control-plane for management and configuration.
- Single pane of glass monitoring, observability and tracing.
- Building and shipping of VMs and containers using an OCI (Open Container Initiative) compliant registry for the artifacts.
- A single technology for support, professional services and completely optional, licensing.
The takeaway, after a decade with Kubernetes, is that we finally have a viable path forward to truly unify virtual machines and containers simply by running the virtual machine in the container. This concept is nothing new and KubeVirt was not first. Weโve had containers running KVM instances long before Kubernetes 1.0.0 and not to wrongly cite any historic references, there is a press release from 2011. Take your guess!
Gradual adoption
Given the nearly two-decade long runway IT practitioners have had with virtual machine management with VMware, there are going to be steep adoption and learning curves. Many vendors focus solely on the vSphere installed base and tailor the packaging to compel organizations to make the leap. Since Broadcom took over, the licensing cost structure for vSphere is cited as the most common denominator in the messaging. Other offers may be in play such as migration tools, free training and consultation.
Do organizations just want to perform a lateral migration of their VM estate without any other operational gain besides being slightly more cost effective? Cost is a strong motivator but learning the ways of Platform Engineering, DevOps, GitOps and idempotent systems management at scale are far more important for a business to adopt and stay ahead on the technology curve.
A few popular patterns to perform VM management with KubeVirt on various Kubernetes distributions are described in Management paradigms for virtual machines running on Kubernetes.
Ubiquitous Kubernetes and KVM
Without knowing, we woke up one day and collectively decided that Kubernetes is now boring. Weโre past the ten-year anniversary and adoption is flourishing. We can use more industry jargon claiming no one will get fired for deploying Kubernetes for containerized workloads (donโt deploy on Fridays).
KVM is certainly on everyoneโs mind these days when we discuss hypervisor alternatives. Many partner solutions rely on KVM for the hypervisor part, including HPE Morpheus VM Essentials Software. KVM provides all the right abstractions, performance and feature set to build robust products for the future, again, based on the proven track record of many of HPEโs partners.
KVM has been around in the mainline kernel since 2007 and yours truly have been using KVM as the daily driver for the last ten or so years. When your ability is best utilized for the actual workloads that matter for the business you support, itโs easy to just dumb down the hypervisor part to what it is, a means to an end. I just need to churn VMs to run my Kubernetes clusters on a bunch of servers, now with the benefit that I can run these VMs on bare-metal Kubernetes to deploy and manage other Kubernetes clusters.
The ubiquity of Kubernetes and KVM is now riper than ever for everyone to take advantage of.
HPE ecosystem for success
The HPE server, networking and storage portfolio provides the most comprehensive building blocks for private and hybrid cloud Kubernetes deployments. Whether gearing towards both traditional workloads on Intel/AMD and modern AI workloads running on ARM architecture. Aruba will provide the data center networking fabric while HPE Alletra Storage MP B10000 provides block and file storage and HPE Alletra Storage MP X10000 serve object storage to Kubernetes workloads.
A vital component of any mission-critical VM deployment is the storage infrastructure. KubeVirt is no different. The key element that customers and partners are looking for is Container Storage Interface (CSI) drivers that provide the ReadWriteMany capability for raw block volumes along with data management features such as snapshot and cloning to provide live migration of VMs and efficient cloning of VM templates. The HPE CSI Driver for Kubernetes supports HPE Alletra Storage MP B10000 and the HPE Alletra 5000/6000/9000 family which integrates seamlessly with partner Kubernetes distributions that supports KubeVirt. A recent technology blog on Migrating vSphere Virtual Volumes to KubeVirt with HPE CSI Driver for Kubernetes demonstrates how users can seamlessly start to ease the migration off vSphere.
- Red Hat OpenShift Virtualization offer customers a smooth experience for running VMs and containers on HPE infrastructure and may be consumed through HPE GreenLake. The HPE CSI Operator for OpenShift is a certified CSI driver that offers a wealth of features to enable a wide range of use cases from edge to cloud.
- SUSE Virtualization offers a Hyper-Converged Infrastructure (HCI) solution that since v1.5.0 can be expanded to grow into a Converged Infrastructure (CI) solution using HPE Alletra Storage MP B10000 for storage and data management of VMs using the HPE CSI Driver for Kubernetes.
- Canonical Kubernetes is uniquely positioned to offer enterprise customers Kubernetes Long Term Support (LTS) and Day One support for upstream Kubernetes. With a comprehensive portfolio of Kubernetes runtimes, such as MicroK8s and Kubernetes Charmed, itโs an easy way to learn about KubeVirt on HPE certified servers and storage platforms.
The KubeVirt Operator may be installed on any supported version of Kubernetes, bare-metal or VMs (with host CPU passthrough enabled), along with the HPE CSI Driver for Kubernetes to begin your KubeVirt journey today.
Summary
Start exploring virtual machines and containers on Kubernetes with KubeVirt today. Experience the workload unification on heterogenous cloud native infrastructure delivered by HPE.
- Learn how to Turbo charge your data and workloads with cloud native storage.
- Explore Red Hat OpenShift Virtualization.
- Interested in Kubernetes LTS? Canonical Kubernetes is the answer.
- SUSE Virtualization (formerly Harvester) offers an HCI stack optimized for VM management with KubeVirt.
Stay tuned to Around The Storage Block for all things persistent storage for Kubernetes with HPE Alletra Storage MP B10000 and neighboring ecosystems. Discuss Kubernetes and KubeVirt on the HPE Developer Community Slack community with fellow technologists. Sign up on developer.hpe.com/slack-signup.
And last, but not least, donโt forget the blogs embedded in the blog regarding management and migrating of VMs and data.
ยน = https://www.gartner.com/en/documents/5748515 via Red Hat partnership
MichaelMattsson
Data & Storage Nerd, Containers, DevOps, IT Automation
