<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Siddhi Chavan on Medium]]></title>
        <description><![CDATA[Stories by Siddhi Chavan on Medium]]></description>
        <link>https://medium.com/@siddhic24?source=rss-f77915d403e9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*yTRrdb0SPXE2mrgYBx0Ixw.jpeg</url>
            <title>Stories by Siddhi Chavan on Medium</title>
            <link>https://medium.com/@siddhic24?source=rss-f77915d403e9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 03:47:23 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@siddhic24/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Breaking the GPU Bottleneck: Multi-Tenant AI Infrastructure with IBM Fusion HCI and NVIDIA vGPU]]></title>
            <link>https://medium.com/@siddhic24/breaking-the-gpu-bottleneck-multi-tenant-ai-infrastructure-with-ibm-fusion-hci-and-nvidia-vgpu-ca3ba29fd862?source=rss-f77915d403e9------2</link>
            <guid isPermaLink="false">https://medium.com/p/ca3ba29fd862</guid>
            <dc:creator><![CDATA[Siddhi Chavan]]></dc:creator>
            <pubDate>Tue, 17 Mar 2026 13:49:45 GMT</pubDate>
            <atom:updated>2026-03-17T13:49:45.072Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FQ90XhK-CnUaAsM1IFQBKg.jpeg" /></figure><h3><strong>The Enterprise AI Infrastructure Challenge</strong></h3><p>As organizations accelerate their AI initiatives, they face a critical infrastructure dilemma: GPU resources are expensive, often underutilized, and difficult to share across multiple workloads. Traditional approaches like PCI passthrough lock entire GPUs to single virtual machines, creating resource silos and limiting flexibility. Meanwhile, AI teams compete for scarce GPU access, slowing innovation and inflating costs.</p><h3><strong>IBM Fusion HCI: Purpose-Built for Modern AI Workloads</strong></h3><p>IBM Fusion HCI provides a fully integrated OpenShift appliance, bringing together compute, networking, storage, GPU, and data resiliency. Unlike traditional virtualization platforms, Fusion HCI enables organizations to:</p><ul><li><strong>Run containerized and VM workloads side-by-side</strong> on a single platform, eliminating infrastructure fragmentation</li><li><strong>Manage VMs as Kubernetes-native resources</strong> through OpenShift Virtualization, bringing cloud-native agility to traditional workloads</li><li><strong>Maximize GPU utilization</strong> through NVIDIA vGPU technology, partitioning physical GPUs into multiple virtual instances with dedicated memory and compute</li><li><strong>Scale AI infrastructure efficiently</strong> without the overhead of separate virtualization stacks</li></ul><h3>Why vGPU Changes the Game</h3><p>NVIDIA vGPU technology transforms how enterprises consume GPU resources. Instead of dedicating an entire physical GPU to a single VM (PCI passthrough), vGPU enables:</p><p>✅ <strong>Multi-tenancy</strong>: One physical GPU serves multiple VMs simultaneously</p><p>✅ <strong>Flexible allocation</strong>: Right-size GPU resources per workload (e.g., 2 VMs per GPU with DC-48Q profile)</p><p>✅ <strong>Better ROI</strong>: Increase GPU utilization from typical 20–30% to 70%+</p><p>✅ <strong>Faster provisioning</strong>: Spin up GPU-enabled VMs in minutes, not days</p><p>✅ <strong>Workload isolation</strong>: Each VM gets dedicated GPU memory and compute</p><p>Combined with IBM Fusion HCI’s enterprise-grade capabilities and the NVIDIA GPU Operator’s automation, this solution delivers production-ready AI infrastructure that scales with your business needs.</p><h3>Real-World Validation: GPT-OSS Inference</h3><p>This guide demonstrates the complete implementation from infrastructure setup through running GPT-OSS, a real-world large language model, on a vGPU-enabled VM. You’ll see how IBM Fusion HCI makes enterprise AI both accessible and cost-effective.</p><blockquote><strong>Key steps covered:</strong></blockquote><ol><li>Prerequisites</li><li>Building the vGPU Manager container image with the NVIDIA vGPU software</li><li>Enabling IOMMU driver on host</li><li>Label worker node</li><li>Create secret in nvidia-gpu-operator namespace</li><li>Creating a ClusterPolicy to enable vGPU manager in vm-vgpu mode</li><li>Labeling the GPU node with the appropriate vGPU profile</li><li>Configuring OpenShift Virtualization (HyperConverged CR) to permit vGPU mediated devices</li><li>Creating an OpenShift Virtualization VM with a vGPU device assigned</li><li>Installing the NVIDIA vGPU guest driver inside the VM</li><li>Running GPT-OSS as an AI inference workload on the vGPU-enabled VM</li></ol><blockquote><strong>Prerequisites:</strong></blockquote><p><strong>Infrastructure Requirements:</strong></p><ul><li>IBM Fusion HCI cluster up and running</li><li>At least one worker node with a supported NVIDIA GPU card type. In my case I have used RTX PRO 6000 GPU cards.</li></ul><p><strong>Required Operators</strong>:</p><ul><li>NVIDIA GPU Operator</li><li>Node Feature Discovery (NFD) operator with its default CR.</li><li>OpenShift Virtualization operator with its default hyperconverged CR.</li></ul><blockquote><strong>Building the vGPU Manager container image with the NVIDIA vGPU software</strong></blockquote><p>The NVIDIA GPU Operator requires a custom vGPU Manager driver image. This image installs the vGPU host driver on OpenShift worker nodes.</p><p><strong>Best Practice</strong>: Always validate GPU compatibility with vGPU software versions before downloading. For example, NVIDIA RTX Pro 6000 Blackwell requires vGPU Software Platform version 19.0 or higher.</p><ul><li>Log in to the <a href="https://nvid.nvidia.com/">NVIDIA Licensing Portal</a>, navigate to Software Downloads, and download the vGPU Manager package for Linux KVM.</li><li>Once downloaded, Unzip the bundle to obtain the NVIDIA vGPU Manager for Linux file, NVIDIA-Linux-x86_64-&lt;version&gt;-vgpu-kvm.run</li><li>Open a terminal and clone the driver container image repository</li></ul><pre>$ git clone https://github.com/NVIDIA/gpu-driver-container.git<br>$ cd gpu-driver-container</pre><ul><li>Copy the NVIDIA vGPU manager from your extracted ZIP file to the operating system version you want to build the image for:<br>For example: RHEL9</li></ul><pre>$ cp &lt;local-driver-download-directory&gt;/*-vgpu-kvm.run vgpu-manager/rhel9/</pre><ul><li>Set the environment variables</li></ul><pre>For RedHat OpenShift 4.18 or earlier:<br>$ export PRIVATE_REGISTRY=my/private/registry VGPU_HOST_DRIVER_VERSION=&lt;downloaded vgpu software version&gt; OS_TAG=rhcos4.18<br><br>For OpenShift 4.19 and later:<br>$ export PRIVATE_REGISTRY=my/private/registry VGPU_HOST_DRIVER_VERSION=&lt;downloaded vgpu software version&gt; OS_TAG=rhel9.6</pre><ul><li>Build the NVIDIA vGPU Manager image</li></ul><pre>$ VGPU_HOST_DRIVER_VERSION=${VGPU_HOST_DRIVER_VERSION} IMAGE_NAME=${PRIVATE_REGISTRY}/vgpu-manager make build-vgpuhost-${OS_TAG}</pre><ul><li>Push the NVIDIA vGPU Manager image to your private registry</li></ul><pre>$ VGPU_HOST_DRIVER_VERSION=${VGPU_HOST_DRIVER_VERSION} IMAGE_NAME=${PRIVATE_REGISTRY}/vgpu-manager make push-vgpuhost-${OS_TAG}</pre><blockquote><strong>Enabling IOMMU driver on host</strong></blockquote><p>IOMMU (Input-Output Memory Management Unit) is required for secure device passthrough and memory isolation.</p><ul><li>Create 100-worker-kernel-arg-iommu.yaml</li></ul><pre>apiVersion: machineconfiguration.openshift.io/v1<br>kind: MachineConfig<br>metadata:<br>  labels:<br>    machineconfiguration.openshift.io/role: worker<br>  name: 100-worker-iommu<br>spec:<br>  config:<br>    ignition:<br>      version: 3.2.0<br>  kernelArguments:<br>      - intel_iommu=on<br>      # If you are using AMD CPU, include the following argument:<br>      # - amd_iommu=on</pre><ul><li>Deploy MachineConfig object</li></ul><pre>$ oc create -f 100-worker-kernel-arg-iommu.yaml</pre><ul><li>Log in to GPU node’s terminal to verify if kernel argument is enabled using following command</li></ul><pre>$ cat /proc/cmdline<br><br>example output:<br>intel_iommu=on<br>     or<br>amd_iommu=on</pre><blockquote>Label worker node</blockquote><p>Label the GPU-enabled worker node to indicate it should run vGPU workloads:</p><pre>$ oc label node &lt;node-name&gt; --overwrite nvidia.com/gpu.workload.config=vm-vgpu</pre><blockquote><strong>Create Image Pull Secret</strong></blockquote><ul><li>Navigate to <strong>Workloads → Secrets</strong> in OpenShift Console, select nvidia-gpu-operator as your project</li><li>Click <strong>Create → Image Pull Secret</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Axy6tRD7X4OA0gb2QT8r9A.png" /><figcaption>Secret fields</figcaption></figure><ul><li>Fill following values:<br>1. Secret name: private-registry-secret<br>2. Authentication type: Image registry credentials<br>3. Registry server address: &lt;private-registry_address&gt;<br>4. Username: $oauthtoken<br>5. Password: &lt;API-KEY&gt;<br>6. Email: &lt;YOUR-EMAIL&gt;</li><li>Click on create.</li></ul><blockquote><strong>Creating a ClusterPolicy to enable vGPU manager in </strong><strong>vm-vgpu mode</strong></blockquote><p>This configures the NVIDIA GPU Operator to deploy vGPU components across your cluster.</p><ul><li>Navigate to <strong>Installed Operators → NVIDIA GPU Operator → ClusterPolicy</strong></li><li>Click <strong>Create ClusterPolicy</strong></li><li>Once clicked on cluster policy select yaml view and add/update following details:<br>1. vgpuManager.enabled=true<br>2. vgpuManager.repository=&lt;path to private repository&gt;<br>3. vgpuManager.image=vgpu-manager<br>4. vgpuManager.version=&lt;driver version&gt;<br>5. vgpuManager.imagePullSecrets={&lt;name of image pull secret&gt;}<br>6. vgpuDeviceManager.enabled=true<br>7. sandboxWorkloads.enabled=true<br>8. sandboxDevicePlugin.enabled=true<br>9. sandboxWorkloads.defaultWorkload= vm-vgpu<br>10. vfioManager.enabled= false</li><li>Verify and click on create</li></ul><blockquote><strong>Labeling GPU Node with vGPU Profile</strong></blockquote><ul><li>Every NVIDIA GPU has a PCI device ID in the format <strong>&lt;vendor_id&gt;:&lt;device_id&gt;</strong>.<br> For example, the NVIDIA RTX Pro 6000 Blackwell uses the PCI ID <strong>10de:2bb5</strong>, where <strong>10de</strong> is the NVIDIA vendor ID and <strong>2bb5</strong> is the device-specific ID.</li><li><strong>Node Feature Discovery (NFD)</strong> automatically detects PCI devices on each node and exposes them as <strong>node labels</strong>, allowing you to retrieve your GPU’s device ID directly from the node.</li><li>After obtaining the device ID, look it up in the <strong>default-vgpu-devices-config ConfigMap</strong> in the <strong>nvidia-gpu-operator</strong> namespace.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fAcrSqXpEMkNW85FMVSRSg.png" /><figcaption>default configmap</figcaption></figure><ul><li>In this ConfigMap, you can search using the <strong>&lt;device_id&gt;&lt;vendor_id&gt;</strong> combined format (for example <strong>0x2BB510DE</strong>) to find the corresponding <strong>supported vGPU profile names</strong> and the <strong>number of instances available per physical GPU</strong>.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gfQA0J6wfxfvRDfAbTFU7A.png" /></figure><pre>device-filter: &quot;0x2BB510DE&quot;<br>          devices: all<br>          vgpu-devices:<br>            DC-48Q: 2</pre><ul><li>Once found, apply following label</li></ul><pre>$ oc label node &lt;gpu-node-name&gt; \<br>  nvidia.com/vgpu.config=&lt;vgpu-device-name&gt; \<br>  --overwrite<br><br>example,<br>$ oc label node &lt;gpu-node-name&gt; \<br>  nvidia.com/vgpu.config=DC-48Q \<br>  --overwrite</pre><ul><li>Verify all following pods in nvidia-gpu-operator are up and running</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RKzBC46jCA7rxUyhDuc-Og.png" /></figure><blockquote><strong>Configuring OpenShift Virtualization (HyperConverged CR) to permit vGPU mediated devices</strong></blockquote><p>For OpenShift Virtualization to expose vGPU slices to VMs, configure the HyperConverged CR</p><ul><li>Run following command to get resource name of GPU</li></ul><pre>$ oc get node &lt;gpu_node_name&gt; -o json | jq &#39;.status.allocatable | with_entries(select(.key | startswith(&quot;nvidia.com/&quot;))) | with_entries(select(.value != &quot;0&quot;))&#39;<br><br>Example output,<br>{<br>  &quot;nvidia.com/NVIDIA_RTX_Pro_6000_Blackwell_DC-48Q&quot;: &quot;4&quot;<br>}</pre><ul><li>Enable disableMDevConfiguration feature gate</li></ul><pre>$ oc patch hyperconverged kubevirt-hyperconverged -n openshift-cnv \<br>  --type=merge -p &#39;{<br>    &quot;spec&quot;: {<br>      &quot;featureGates&quot;: {<br>        &quot;disableMDevConfiguration&quot;: true<br>      }<br>    }<br>  }&#39;</pre><ul><li>Register the vGPU device under permittedHostDevices</li></ul><pre>$ oc patch hyperconverged kubevirt-hyperconverged -n openshift-cnv \<br>  --type=merge -p &#39;{<br>    &quot;spec&quot;: {<br>      &quot;permittedHostDevices&quot;: {<br>        &quot;mediatedDevices&quot;: [<br>          {<br>            &quot;mdevNameSelector&quot;: &quot;&lt;mdev_device_name&gt;&quot;,<br>            &quot;resourceName&quot;: &quot;&lt;resource_name&gt;&quot;,<br>            &quot;externalResourceProvider&quot;: true<br>          }<br>        ]<br>      }<br>    }<br>  }&#39;<br><br>example,<br>$ oc patch hyperconverged kubevirt-hyperconverged -n openshift-cnv \<br>  --type=merge -p &#39;{<br>    &quot;spec&quot;: {<br>      &quot;permittedHostDevices&quot;: {<br>        &quot;mediatedDevices&quot;: [<br>          {<br>            &quot;mdevNameSelector&quot;: &quot;NVIDIA RTX Pro 6000 Blackwell DC-48Q&quot;,<br>            &quot;resourceName&quot;: &quot;nvidia.com/NVIDIA_RTX_Pro_6000_Blackwell_DC-48Q&quot;,<br>            &quot;externalResourceProvider&quot;: true<br>          }<br>        ]<br>      }<br>    }<br>  }&#39;</pre><p><strong>Note</strong>:</p><ul><li>mdevNameSelector: Human-readable vGPU profile name from NVIDIA driver</li><li>resourceName: Kubernetes resource name exposed by GPU Operator</li><li>externalResourceProvider: true: Indicates GPU Operator manages this resource</li></ul><blockquote><strong>Creating a VM with vGPU Assignment</strong></blockquote><p>This step demonstrates how quickly teams can provision GPU-enabled VMs minutes instead of days with traditional infrastructure.</p><ul><li>Navigate to <strong>Virtualization → VirtualMachines</strong> in OpenShift Console</li><li>Click <strong>Create → From template</strong></li><li>Select <strong>Red Hat Enterprise Linux 9 VM</strong></li><li>Click <strong>Customize VirtualMachine</strong></li><li>Go to <strong>Overview → Hardware devices → GPU devices</strong></li><li>Select your vGPU devices from the dropdown</li><li>Click <strong>Save</strong> then <strong>Create VirtualMachine</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9TgOLVk5F63i1GUd6tPDqw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VDXL8c6mGnXqMi_OBK8Bsg.png" /></figure><blockquote><strong>Installing NVIDIA vGPU Guest Driver</strong></blockquote><p>The guest driver enables the VM to communicate with the vGPU slice allocated to it.</p><ul><li>Transfer the guest driver to the VM:<br>1. In <strong>Step 1</strong>, you downloaded the <strong>NVIDIA vGPU Linux KVM software package</strong> from the NVIDIA Licensing Portal.<br>2. Navigate to the <strong>Guest_Drivers</strong> directory inside the extracted folder.<br>3. In this directory, you will find the guest driver file named <strong>NVIDIA-Linux-x86_64-&lt;version&gt;-grid.run</strong> (for example, NVIDIA-Linux-x86_64-580.126.09-grid.run).<br>4. Transfer this file to your VM.</li></ul><p>Note: The guest driver version must match the vGPU Manager version installed on the host.</p><ul><li>Copy the driver into the VM using a HTTP server on your local machine</li></ul><pre># On your local machine — serve the driver file<br>$ python3 -m http.server 8080 --directory /path/to/driver/<br><br># Inside the VM — download it<br>$ curl -O http://&lt;your-local-ip&gt;:8080/NVIDIA-Linux-x86_64-&lt;version&gt;-grid.run</pre><ul><li>Disable Secure Boot (if not already done)<br>The NVIDIA guest driver is an unsigned kernel module and will fail to load if Secure Boot is enabled in the VM. Ensure the VM spec has Secure Boot disabled:</li></ul><pre>firmware:<br>  bootloader:<br>    efi:<br>      secureBoot: false</pre><ul><li>Setup Subscription Manager and enable following repos</li></ul><pre>$ sudo subscription-manager register<br>$ sudo subscription-manager repos \<br>  --enable=rhel-9-for-x86_64-baseos-rpms \<br>  --enable=rhel-9-for-x86_64-appstream-rpms</pre><ul><li>Install required packages</li></ul><pre>$ sudo dnf install -y kernel kernel-devel kernel-headers gcc make elfutils-libelf-devel</pre><ul><li>Check kernel version matches kernel-devel</li></ul><pre>$ uname -r<br>$ rpm -q kernel-devel</pre><p>note: If versions don’t match, update kernel using following command:</p><pre>$ sudo dnf update -y kernel kernel-devel kernel-headers<br>$ sudo reboot</pre><ul><li>After reboot, disable nouveau driver</li></ul><pre>$ sudo bash -c &quot;echo blacklist nouveau &gt; /etc/modprobe.d/blacklist-nouveau.conf&quot;<br>$ sudo bash -c &quot;echo options nouveau modeset=0 &gt;&gt; /etc/modprobe.d/blacklist-nouveau.conf&quot;<br>$ sudo dracut --force<br>$ sudo reboot</pre><ul><li>Make driver executable and install</li></ul><pre>$ chmod +x NVIDIA-Linux-x86_64–&lt;version&gt;-grid.run<br>$ sudo ./NVIDIA-Linux-x86_64–&lt;version&gt;-grid.run</pre><ul><li>Verify Installation</li></ul><pre>$ nvidia-smi</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tm9EjULA8NujaQ9o1Y2VnQ.png" /></figure><blockquote><strong>Running GPT-OSS as an AI inference workload on the vGPU-enabled VM</strong></blockquote><p>With the vGPU guest driver installed and nvidia-smi confirming GPU access, we can now run an AI inference workload directly inside the VM. We use Ollama a lightweight runtime for running large language models locally to serve GPT-OSS on the vGPU.</p><ul><li>Install Ollama inside the VM</li></ul><pre>$ curl -fsSL https://ollama.com/install.sh | sh</pre><ul><li>Run GPT-OSS</li></ul><pre>$ ollama run gpt-oss</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*Cwuy6M2SIpjS2d_l54qj4Q.gif" /><figcaption>Demonstration of GPU consumption while running gpt-oss model</figcaption></figure><p><strong>What This Demonstrates</strong>: A production-ready large language model running on a virtualized GPU slice proving that vGPU delivers real AI performance without dedicated hardware.</p><p>By implementing NVIDIA vGPU on IBM Fusion HCI, organizations achieve:</p><p>✅ <strong>3–5x GPU utilization improvement</strong> compared to PCI passthrough</p><p>✅ <strong>Faster AI project deployment</strong> with self-service VM provisioning</p><p>✅ <strong>Lower infrastructure costs</strong> by maximizing existing GPU investments</p><p>✅ <strong>Simplified operations</strong> with unified Kubernetes-native management</p><p>✅ <strong>Production-ready AI infrastructure</strong> that scales with business needs</p><blockquote><strong>Best Practices Summary</strong></blockquote><ol><li><strong>Always validate GPU compatibility</strong> with vGPU software versions before deployment</li><li><strong>Match guest driver versions</strong> to vGPU Manager versions exactly</li><li><strong>Use appropriate vGPU profiles</strong> based on workload memory requirements</li><li><strong>Monitor GPU utilization</strong> to optimize profile allocation over time</li></ol><p>Ready to maximize your GPU investment? Start with IBM Fusion HCI and NVIDIA vGPU today!!</p><p>Thank you for taking the time to read this blog. I appreciate your interest and hope you found the information helpful and easy to follow. Your engagement is invaluable, and I encourage you to reach out with any questions or feedback. Your input helps improve future content and ensures we continue to provide useful and relevant information.</p><p>Resources: <br><a href="https://docs.nvidia.com/datacenter/cloud-native/openshift/latest/openshift-virtualization.html">https://docs.nvidia.com/datacenter/cloud-native/openshift/latest/openshift-virtualization.html</a><br><a href="https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/virtualization/managing-vms#virt-configuring-virtual-gpus">https://docs.redhat.com/en/documentation/openshift_container_platform/4.18/html/virtualization/managing-vms#virt-configuring-virtual-gpus</a></p><p>𝐃𝐢𝐬𝐜𝐥𝐚𝐢𝐦𝐞𝐫: The views and opinions expressed in this guide are solely my own and do not reflect those of IBM. This is shared for informational purposes only and should not be considered as official guidance or advice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ca3ba29fd862" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deployment of Bare Metal Hosted Control Plane within a Disconnected IBM Fusion HCI Environment]]></title>
            <link>https://medium.com/@siddhic24/deployment-of-bare-metal-hosted-control-plane-within-a-disconnected-ibm-fusion-hci-environment-36dc4e2168d7?source=rss-f77915d403e9------2</link>
            <guid isPermaLink="false">https://medium.com/p/36dc4e2168d7</guid>
            <dc:creator><![CDATA[Siddhi Chavan]]></dc:creator>
            <pubDate>Fri, 09 Jan 2026 04:56:59 GMT</pubDate>
            <atom:updated>2026-10-29T13:12:06.913Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7f-XYhspmGPTw0aPnkjRCQ.jpeg" /><figcaption>Source: <a href="https://www.linkedin.com/in/sahil-tah/">https://www.linkedin.com/in/sahil-tah/</a></figcaption></figure><blockquote>Introduction:</blockquote><p>As organizations expand OpenShift adoption across teams, they must efficiently share infrastructure while maintaining strong isolation and performance. Platform teams need a scalable approach that avoids hardware sprawl and operational overhead.</p><p>IBM Fusion HCI provides a fully integrated OpenShift appliance, bringing together compute, networking, storage, GPU, and data resiliency. With Hosted Control Plane (HCP), multiple OpenShift clusters can run on a single Fusion HCI system, enabling teams to operate independently while maximizing infrastructure utilization.</p><p>Many enterprises also face strict security and compliance requirements that limit access to public registries or external networks. In such environments, OpenShift must run in a fully disconnected model, often with demands for higher performance and tighter isolation.</p><p>To address this, IBM Fusion HCI supports <strong>Bare Metal Hosted Control Plane deployments in disconnected environments</strong>, allowing control planes to run directly on bare metal without external connectivity — ideal for air-gapped and mission-critical workloads.</p><p>In this blog, we’ll walk through setting up a bare metal HCP deployment in a disconnected environment and highlight how it delivers simplicity, resilience, and optimal performance.</p><blockquote>Key steps covered:</blockquote><ol><li>Prerequisites.</li><li>Prepare Hub Cluster of Bare Metal HCP deployment.<br> 2.1. Network Configuration.<br> 2.2. Configure web server.<br> 2.3. Configuring image mirroring.<br> 2.4. Apply objects (IDMS/ICSP and Catalog Sources) in hub cluster.<br> 2.5. Create custom-registry config map.<br> 2.6. Deploy AgentServiceConfig resource.<br> 2.7. Configure TLS certificates.</li><li>Create Bare Metal HCP cluster.</li><li>Create NMStateConfig per host.</li><li>Add host with Discovery ISO.</li><li>Approve hosts added in host inventory.</li><li>Label agents.</li><li>Configure ingress.</li></ol><blockquote>Prerequisites:</blockquote><p>1. Multicluster engine for Kubernetes (MCE) Operator with its multiclusterengine CR.</p><p>2. MetalLB operator with its 3 CRs: metallb, IPAddressPool and L2Advertisement.</p><blockquote><strong>Prepare Hub Cluster of Bare Metal HCP deployment</strong></blockquote><blockquote>Network Configuration:</blockquote><p>Before starting the deployment, make sure to create DNS entries for:</p><ul><li>Host that will be used as a node in the Bare Metal HCP cluster deployment.</li><li>Ingress. (This IP address is used to create a load balancer on the Hosted Control Plane cluster to allow ingress to that cluster.)</li></ul><blockquote>Configure web server:</blockquote><p>You need to configure an additional web server to host the Red Hat Enterprise Linux CoreOS (RHCOS) images that are associated with the OpenShift Container Platform release. In my case, I had IBM Fusion HCI with service node which has additional web server.</p><p>Procedure to upload ROOTFS_IMG_URL and LIVE_ISO_URL image over web server:</p><ul><li>Extract an openshift-install from OCP release using following command (run commands from bastion node that has connectivity to public registry):</li></ul><pre>oc adm -a ${LOCAL_SECRET_JSON} release extract --command=openshift-install \<br>&quot;${LOCAL_REGISTRY}/${LOCAL_REPOSITORY}:${OCP_RELEASE}-${ARCHITECTURE}&quot;</pre><p>After running this command, you will see openshift-install in <br> current directory.</p><ul><li>Get ROOTFS_IMG_URL using installer extracted in step 1 and download file using following commands:</li></ul><pre>ROOTFS_IMG_URL=&quot;$(./openshift-install coreos print-stream-json | jq -r &#39;.architectures.x86_64.artifacts.metal.formats.pxe.rootfs.location’)”          <br><br>wget $ROOTFS_IMG_URL --no-check-certificate </pre><ul><li>Get LIVE_ISO_URL using installer extracted in step A and download file using following commands:</li></ul><pre>LIVE_ISO_URL=&quot;$(./openshift-install coreos print-stream-json | jq -r &#39;.architectures.x86_64.artifacts.metal.formats.iso.disk.location’)”         <br><br>wget $LIVE_ISO_URL --no-check-certificate    </pre><ul><li>Copy both files to web server:<br>log in to the service node as a KNI user, create the directory /var/www/html/hcp/images and copy the extracted files into this location.</li></ul><blockquote>Configuring image mirroring:</blockquote><p>Image mirroring is the process of fetching images from external registries, such as registry.redhat.com or quay.io, and storing them in your private registry.</p><p>In the following procedures, the oc-mirror tool is used, which is a binary that uses the ImageSetConfiguration object.</p><p>In the file, you can specify the following information:</p><p>- The OpenShift Container Platform versions to mirror. The <br> versions are in quay.io.<br>- The additional Operators to mirror. Select packages individually.<br>- The extra images that you want to add to the repository.</p><p>Example:</p><pre>apiVersion: mirror.openshift.io/v2alpha1<br>kind: ImageSetConfiguration<br>mirror:<br>  operators:<br>    - catalog: registry.redhat.io/redhat/redhat-operator-index:v4.18<br>      packages:<br>        - name: &quot;local-storage-operator&quot;<br>        - name: &quot;lvms-operator&quot;<br>        - name: &quot;kubernetes-nmstate-operator&quot;<br>        - name: &quot;redhat-oadp-operator&quot;<br>        - name: &quot;amq-streams&quot;<br>        - name: &quot;kubevirt-hyperconverged&quot;<br>        - name: &quot;metallb-operator&quot;<br>        - name: &quot;multicluster-engine&quot;<br>      full: true<br>  platform:<br>    channels:<br>    - name: stable-4.18<br>      minVersion: 4.18.10<br>      maxVersion: 4.18.10<br>      type: ocp<br>    kubeVirtContainer: true<br>    graph: true</pre><blockquote>Apply objects (IDMS/ICSP and Catalog Sources) in hub cluster:</blockquote><p>After the mirroring process is complete, you need to apply two objects in the management cluster:</p><p>- ImageContentSourcePolicy (ICSP) or ImageDigestMirrorSet (IDMS)<br>- Catalog sources</p><p>Once done, simply install the operators needed from operator hub and create all necessary custom resources which are mentioned in prerequisites section.</p><blockquote>Create custom-registry config map:</blockquote><p>To tailor the deployment for a disconnected environment, define a ConfigMap in multicluster-engine namespace that specifies the custom registry details</p><p>Example:</p><pre>kind: ConfigMap<br>apiVersion: v1<br>metadata:<br>  name: custom-registries<br>  namespace: multicluster-engine<br>data:<br>  &lt;registry_name&gt;..&lt;port&gt;: |  <br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE-----<br>  &lt;registry_name&gt;..&lt;port&gt;: |<br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE----<br><br>  registries.conf: |<br>    unqualified-search-registries = [&quot;registry.access.redhat.com&quot;, &quot;docker.io&quot;]<br>    short-name-mode = &quot;&quot;<br>    [[registry]]<br>       prefix = &quot;&quot;<br>       location = &quot;quay.io/openshift-release-dev/ocp-v4.0-art-dev&quot;<br>       mirror-by-digest-only = true<br>       [[registry.mirror]]<br>         location = &quot;$TARGET_PATH/openshift/release&quot;<br><br>    [[registry]]<br>       prefix = &quot;&quot;<br>       location = &quot;quay.io/openshift-release-dev/ocp-release&quot;<br>       mirror-by-digest-only = true<br>       [[registry.mirror]]<br>         location = &quot;$TARGET_PATH/openshift/release-images&quot;<br><br>    [[registry]]<br>       prefix = &quot;&quot;<br>       location = &quot;registry.redhat.io/multicluster-engine&quot;<br>       mirror-by-digest-only = true<br>       [[registry.mirror]]<br>         location = &quot;$TARGET_PATH/multicluster-engine&quot;<br><br>    [[registry]]<br>       prefix = &quot;&quot;<br>       location = &quot;registry.redhat.io/openshift4&quot;<br>       mirror-by-digest-only = true<br>       [[registry.mirror]]<br>         location = &quot;$TARGET_PATH/openshift4&quot;<br><br>    [[registry]]<br>       prefix = &quot;&quot;<br>       location = &quot;registry.redhat.io/rhel9&quot;<br>       mirror-by-digest-only = true<br>       [[registry.mirror]]<br>         location = &quot;$TARGET_PATH/rhel9&quot;<br><br>    [[registry]]<br>       prefix = &quot;&quot;<br>       location = &quot;registry.redhat.io/rhel8&quot;<br>       mirror-by-digest-only = true<br>       [[registry.mirror]]<br>         location = &quot;$TARGET_PATH/rhel8&quot;<br><br>    [[registry]]<br>       prefix = &quot;&quot;<br>       location = &quot;registry.redhat.io/redhat&quot;<br>       mirror-by-digest-only = true<br>       [[registry.mirror]]<br>         location = &quot;$TARGET_PATH/redhat&quot;</pre><blockquote>Deploy AgentServiceConfig resource:</blockquote><p>The <strong>AgentServiceConfig</strong> custom resource, part of the <strong>Assisted Service</strong> add-on in the <strong>multicluster engine Operator</strong>, enables bare metal cluster deployment. After enabling the add-on, deploy the <strong>AgentServiceConfig</strong> to configure the Assisted Service, as shown below:</p><pre>apiVersion: agent-install.openshift.io/v1beta1<br>kind: AgentServiceConfig<br>metadata:<br>  annotations:<br>    unsupported.agent-install.openshift.io/assisted-service-configmap: custom-registries<br>  name: agent<br>spec:<br>  databaseStorage:<br>    accessModes:<br>      - ReadWriteOnce<br>    resources:<br>      requests:<br>        storage: 200Gi<br>    storageClassName: ocs-storagecluster-ceph-rbd<br>  filesystemStorage:<br>    accessModes:<br>      - ReadWriteOnce<br>    resources:<br>      requests:<br>        storage: 200Gi<br>    storageClassName: ocs-storagecluster-ceph-rbd<br>  imageStorage:<br>    accessModes:<br>      - ReadWriteOnce<br>    resources:<br>      requests:<br>        storage: 100Gi<br>  mirrorRegistryRef:<br>    name: custom-registries<br>  osImages:<br>    - cpuArchitecture: x86_64<br>      openshiftVersion: 4.18.20<br>      rootFSUrl: &#39;http://172.20.x.y:&lt;port&gt;/hcp/images/rhcos-418.94.202501221327-0-live-rootfs.x86_64.img&#39;<br>      url: &#39;http://172.20.x.y:&lt;port&gt;/hcp/images/rhcos-418.94.202501221327-0-live.x86_64.iso&#39;<br>      version: 418.94.202501221327-0</pre><blockquote>Configure TLS certificates:</blockquote><p>To enable proper operation in a disconnected deployment, configure the registry CA certificates on the management cluster.</p><ul><li>Add a registry CA to the hub cluster<br>Create a config map in the openshift-config namespace.</li></ul><pre>apiVersion: v1<br>kind: ConfigMap<br>metadata:<br>  name: &lt;config_map_name&gt; <br>  namespace: &lt;config_map_namespace&gt; <br>data: <br>  &lt;registry_name&gt;..&lt;port&gt;: | <br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE-----<br>  &lt;registry_name&gt;..&lt;port&gt;: |<br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE-----<br>  </pre><ul><li>Update the cluster-wide image.config.openshift.io object by applying the following specification:</li></ul><pre>spec:<br>  additionalTrustedCA:<br>    - name: &lt;registry-config-config_map_name&gt;</pre><blockquote>Create Bare Metal HCP cluster:</blockquote><p>A <strong>hosted cluster</strong> in <strong>OpenShift Container Platform</strong> runs its control plane and API endpoint on a management cluster, while including both the control plane and its associated data plane</p><ul><li>Create two namespaces: &lt;cluster-name&gt; and &lt;cluster_namespace-cluster-name&gt; <br>Example: If my cluster name is fusion-bm, 2 namespaces will be<br>1. fusion-bm<br>2. fusion-bm-fusion-bm</li></ul><pre>apiVersion: v1<br>kind: Namespace<br>metadata:<br>  creationTimestamp: null<br>  name: fusion-bm-fusion-bm<br>spec: {}<br>status: {}<br>---<br>apiVersion: v1<br>kind: Namespace<br>metadata:<br>  creationTimestamp: null<br>  name: fusion-bm<br>spec: {}<br>status: {}</pre><ul><li>Create a ConfigMap which contains registry CA certificates in hosted-cluster-namespace<br>Example,</li></ul><pre>kind: ConfigMap<br>apiVersion: v1<br>metadata:<br>  name: user-ca-bundle<br>  namespace: &lt;hosted-cluster-namespace example, fusion-bm&gt;<br>data:<br>  ca-bundle.crt: |<br>    // Registry1 CA<br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE-----<br><br>    // Registry2 CA<br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE-----</pre><ul><li>Create a pull secret and ssh-key in the hosted-cluster-namespace<br>Example,</li></ul><pre>apiVersion: v1<br>kind: Secret<br>metadata:<br>  name: fusion-bm-pull-secret<br>  namespace: &lt;hosted-cluster-namespace example, fusion-bm&gt;<br>data:<br>  .dockerconfigjson: xxxxxxxxx<br>type: kubernetes.io/dockerconfigjson<br>---<br>apiVersion: v1<br>kind: Secret<br>metadata:<br>  name: sshkey-cluster-fusion-bm<br>  namespace: &lt;hosted-cluster-namespace example, fusion-bm&gt;<br>data:<br>  id_rsa.pub: xxxxxxxxx<br>type: Opaque</pre><ul><li>Define a YAML file with the required <strong>RBAC roles</strong> to ensure <strong>Assisted Service agents</strong> in the same <strong>HostedControlPlane namespace</strong> as the hosted control plane can be managed by the <strong>Cluster API<br></strong>Example,</li></ul><pre>apiVersion: rbac.authorization.k8s.io/v1<br>kind: Role<br>metadata:<br>  name: capi-provider-role<br>  namespace: &lt;cluster_namespace-cluster-name example, fusion-bm-fusion-bm&gt;<br>rules:<br>  - verbs:<br>      - &#39;*&#39;<br>    apiGroups:<br>      - agent-install.openshift.io<br>    resources:<br>      - agents</pre><ul><li>Create HostedCluster CR with appropriate values.<br>Example,</li></ul><pre>apiVersion: hypershift.openshift.io/v1beta1<br>kind: HostedCluster<br>metadata:<br>  name: &lt;cluster name example. fusion-bm&gt;<br>  namespace: &lt;hosted-cluster-namespace example, fusion-bm&gt;<br>spec:<br>  fips: false<br>  release:<br>    image: &lt;$TARGET_PATH&gt;/openshift/release-images:&lt;ocp_version&gt;-x86_64<br>  dns:<br>    baseDomain: &lt;xyz.com&gt;<br>  controllerAvailabilityPolicy: HighlyAvailable<br>  etcd:<br>    managed:<br>      storage:<br>        persistentVolume:<br>          size: 8Gi<br>          storageClassName: ocs-storagecluster-ceph-rbd<br>        type: PersistentVolume<br>    managementType: Managed<br>  infrastructureAvailabilityPolicy: SingleReplica<br>  platform:<br>    agent:<br>      agentNamespace: &lt;cluster_namespace-cluster-name example, fusion-bm-fusion-bm&gt;<br>    type: Agent<br>  additionalTrustBundle:<br>    name: user-ca-bundle<br>  networking:<br>    clusterNetwork:<br>      - cidr: 10.132.0.0/14<br>    networkType: OVNKubernetes<br>    serviceNetwork:<br>      - cidr: 172.31.0.0/16<br>  pullSecret:<br>    name: fusion-bm-pull-secret<br>  configuration:<br>    operatorhub:<br>      disableAllDefaultSources: true<br>  capabilities: {}<br>  sshKey:<br>    name: sshkey-cluster-fusion-bm<br>  autoscaling: {}<br>  imageContentSources:<br>    - mirrors:<br>        - $TARGET_PATH/openshift/release<br>      source: quay.io/openshift-release-dev/ocp-v4.0-art-dev<br>    - mirrors:<br>        - $TARGET_PATH/redhat<br>      source: registry.redhat.io/redhat<br>    - mirrors:<br>        - $TARGET_PATH/rhel9<br>      source: registry.redhat.io/rhel9<br>    - mirrors:<br>        - $TARGET_PATH/rhel8<br>      source: registry.redhat.io/rhel8<br>    - mirrors:<br>        - $TARGET_PATH/openshift/release-images<br>      source: quay.io/openshift-release-dev/ocp-release<br>    - mirrors:<br>        - $TARGET_PATH/openshift4<br>      source: registry.redhat.io/openshift4<br>    - mirrors:<br>        - $TARGET_PATH/multicluster-engine<br>      source: registry.redhat.io/multicluster-engine<br>  olmCatalogPlacement: management<br>  services:<br>    - service: APIServer<br>      servicePublishingStrategy:<br>        type: LoadBalancer<br>    - service: Ignition<br>      servicePublishingStrategy:<br>        type: Route<br>    - service: Konnectivity<br>      servicePublishingStrategy:<br>        type: Route<br>    - service: OAuthServer<br>      servicePublishingStrategy:<br>        type: Route<br>   </pre><p>Note: While creating HostedCluster CR, please remove OIDC service.</p><ul><li>Create Nodepool CR<br>Example,</li></ul><pre>apiVersion: hypershift.openshift.io/v1beta1<br>kind: NodePool<br>metadata:<br>  name: fusion-bm-np<br>  namespace: &lt;hosted-cluster-namespace example, fusion-bm&gt;<br>spec:<br>  arch: amd64<br>  clusterName: &lt;hosted cluster name example, fusion-bm&gt;<br>  management:<br>    autoRepair: false<br>    replace:<br>      rollingUpdate:<br>        maxSurge: 1<br>        maxUnavailable: 0<br>      strategy: RollingUpdate<br>    upgradeType: InPlace<br>  nodeDrainTimeout: 0s<br>  platform:<br>    agent:<br>      agentLabelSelector:<br>        matchLabels:<br>          inventory.agent-install.openshift.io/hcp: fusion-bm<br>    type: Agent<br>  release:<br>    image: &#39;$TARGET_PATH/openshift/release-images:&lt;ocp_version&gt;-x86_64&#39;<br>  replicas: &lt;no of nodes. example: 3&gt;</pre><ul><li>Create InfraEnv resource<br>The <strong>InfraEnv</strong> resource in <strong>Assisted Service</strong> holds details like <strong>pullSecretRef</strong> and <strong>sshAuthorizedKey</strong>, used to create a customized <strong>RHCOS</strong> boot image for the hosted cluster.<br>Procedure:<br>1. Create a secret named pull secret in the &lt;hosted-cluster-namespace&gt;-&lt;hosted_cluster_name&gt; namespace<br>Example,</li></ul><pre>apiVersion: v1<br>kind: Secret<br>metadata:<br>  name: pullsecret-fusion-bm<br>  namespace: &lt;cluster_namespace-cluster-name example, fusion-bm-fusion-bm&gt;<br>data:<br>  .dockerconfigjson: xxxxxxxx<br>type: kubernetes.io/dockerconfigjson</pre><ul><li>2. Create an InfraEnv CR in the &lt;hosted-cluster-namespace&gt;-&lt;hosted_cluster_name&gt; namespace.<br>Example,</li></ul><pre>apiVersion: agent-install.openshift.io/v1beta1<br>kind: InfraEnv<br>metadata:<br>  name: fusion-bm<br>  namespace: &lt;cluster_namespace-cluster-name example, fusion-bm-fusion-bm&gt;<br>  labels:<br>    agentclusterinstalls.extensions.hive.openshift.io/location: us<br>    networkType: static<br>spec:<br>  agentLabels:<br>    agentclusterinstalls.extensions.hive.openshift.io/location: us<br>  imageType: minimal-iso<br>  mirrorRegistryRef:<br>    name: custom-registries<br>    namespace: multicluster-engine<br>  ipxeScriptType: DiscoveryImageAlways<br>  cpuArchitecture: x86_64<br>  sshAuthorizedKey: xxxxxxxx<br>  pullSecretRef:<br>    name: pullsecret-fusion-bm<br>  nmStateConfigLabelSelector:<br>    matchLabels:<br>      infraenvs.agent-install.openshift.io: fusion-bm</pre><blockquote>Create NMStateConfig per host and Add host with Discovery ISO to host inventory:</blockquote><p>To learn how to create <strong>NMStateConfig</strong> for each host and add hosts to the inventory, refer to the section <em>“</em><strong><em>Create NMStateConfig for Each Host and Add Host to the Host Inventory</em></strong><em>”</em> in the following blog:<br> <a href="https://medium.com/@siddhic24/create-bare-metal-hosted-control-plane-with-dell-hardware-in-ibm-fusion-hci-c4e8be4f5618">Create Bare Metal Hosted Control Plane with Dell Hardware in IBM Fusion HCI</a></p><blockquote>Label agents :</blockquote><p>After the hosts or nodes are available, update the <strong>Agent</strong> custom resource definition by adding the label specified in the <strong>NodePool</strong> and <strong>InfraEnv</strong> resources to each corresponding agent.<br> For instance, if you used the label inventory.agent-install.openshift.io/hcp: fusion-bm, apply it to the appropriate agent.</p><p>After this Bare Metal Hosted Control plane cluster will be available in sometime.</p><blockquote>Configure ingress :</blockquote><p>Configuring <strong>ingress</strong> is required to access the <strong>UI</strong> of the Bare Metal HCP cluster. To set up ingress, refer to the section <em>“Configuring Ingress for Bare Metal. HCP Cluster”</em> in the following blog:<br> <a href="https://medium.com/@siddhic24/create-bare-metal-hosted-control-plane-with-dell-hardware-in-ibm-fusion-hci-c4e8be4f5618">Create Bare Metal Hosted Control Plane with Dell Hardware in IBM Fusion HCI</a></p><p><strong><em>Note:</em></strong><em> Before configuring ingress, ensure that the </em><strong><em>MetalLB Operator</em></strong><em> is mirrored and the </em><strong><em>Red Hat catalog source</em></strong><em> is applied in the Bare Metal HCP cluster using API login.</em></p><p>And just like that, bare metal HCP on IBM Fusion HCI brings high-performance OpenShift to fully disconnected environments. Secure, isolated, and simple exactly what modern enterprise platforms need.</p><p>Interested in how data services are managed on hosted control planes in IBM Fusion’s disconnected environments? <br>Follow along to find out more: <a href="https://medium.com/@spp17081978/managing-data-services-on-hosted-control-planes-in-ibm-fusion-in-disconnected-environment-9d837ccfab1b">https://medium.com/@spp17081978/managing-data-services-on-hosted-control-planes-in-ibm-fusion-in-disconnected-environment-9d837ccfab1b</a></p><p>Resources: <br><a href="https://www.ibm.com/docs/en/fusion-hci-systems/2.12.0?topic=mcfhuhcp-deploying-bare-metal-hosted-control-plane-cluster-in-disconnected-environment">https://www.ibm.com/docs/en/fusion-hci-systems/2.12.0?topic=mcfhuhcp-deploying-bare-metal-hosted-control-plane-cluster-in-disconnected-environment</a></p><p>Thank you for taking the time to read this blog. I appreciate your interest and hope you found the information helpful and easy to follow. Your engagement is invaluable, and I encourage you to reach out with any questions or feedback. Your input helps improve future content and ensures we continue to provide useful and relevant information.</p><p>𝐃𝐢𝐬𝐜𝐥𝐚𝐢𝐦𝐞𝐫: The views and opinions expressed in this guide are solely my own and do not reflect those of IBM. This is shared for informational purposes only and should not be considered as official guidance or advice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=36dc4e2168d7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Virtualized Hosted Control Plane Deployment within a Disconnected IBM Fusion HCI Environment]]></title>
            <link>https://medium.com/@siddhic24/virtualized-hosted-control-plane-deployment-within-a-disconnected-ibm-fusion-hci-environment-6b04b522a3ba?source=rss-f77915d403e9------2</link>
            <guid isPermaLink="false">https://medium.com/p/6b04b522a3ba</guid>
            <dc:creator><![CDATA[Siddhi Chavan]]></dc:creator>
            <pubDate>Fri, 09 Jan 2026 04:53:00 GMT</pubDate>
            <atom:updated>2026-01-10T08:54:30.460Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7f-XYhspmGPTw0aPnkjRCQ.jpeg" /><figcaption>Source: <a href="https://www.linkedin.com/in/sahil-tah/">https://www.linkedin.com/in/sahil-tah/</a></figcaption></figure><blockquote>Introduction:</blockquote><p>As organizations scale their OpenShift platforms, they face a key challenge: how to maximize shared infrastructure while allowing multiple teams to work independently. Platform teams must deliver clusters quickly without over-provisioning hardware or letting one team’s workloads impact another’s. This is where an architecture that balances isolation with efficient resource sharing becomes essential.</p><p>IBM Fusion HCI meets this need with an all-in-one appliance that delivers enterprise-grade OpenShift by integrating compute, networking, storage, GPU, and data resiliency. Its Hosted Control Plane (HCP) capability allows a single Fusion HCI system to be partitioned into multiple OpenShift clusters, enabling teams to share infrastructure efficiently while maintaining strong isolation.</p><p>At the same time, many organizations operate in security-sensitive environments where access to public registries or external connectivity is restricted. For these scenarios, OpenShift must run fully disconnected, without reliance on public services.</p><p>To address this, IBM Fusion HCI supports <strong>virtualized Hosted Control Plane deployments in disconnected environments</strong>. This capability enables secure, offline OpenShift clusters while preserving the flexibility and efficiency of HCP.</p><p>In this blog, we’ll walk through setting up a virtualized HCP environment in a disconnected setting and highlight how it simplifies secure OpenShift deployments.</p><blockquote>Key steps covered:</blockquote><p>1. Prerequisites.<br> 2. Configuring image mirroring for hosted control planes.<br> 3. Apply objects (IDMS/ICSP and Catalog Sources) in hub cluster.<br> 4. Configure TLS certification.<br> 5. Create virtualized cluster.</p><blockquote>Prerequisites:</blockquote><p>1. Multicluster engine for Kubernetes (MCE) Operator with its multiclusterengine CR.</p><p>2. MetalLB operator with its 3 CRs: metallb, IPAddressPool and L2Advertisement.</p><p>3. OpenShift Virtualization Operator with its hyperconverged CR.</p><p>4. Configure Ingress by running following command:</p><pre>oc patch ingresscontroller default -n openshift-ingress-operator --type=json -p &#39;[{ &quot;op&quot;: &quot;add&quot;, &quot;path&quot;: &quot;/spec/routeAdmission&quot;, &quot;value&quot;: { &quot;wildcardPolicy&quot;: &quot;WildcardsAllowed&quot; } }]&#39;</pre><blockquote>Configuring image mirroring for hosted control planes:</blockquote><p>Image mirroring is the process of fetching images from external registries, such as registry.redhat.com or quay.io, and storing them in your private registry.</p><p>In the following procedures, the oc-mirror tool is used, which is a binary that uses the ImageSetConfiguration object.</p><p>In the file, you can specify the following information:</p><p>- The OpenShift Container Platform versions to mirror. The <br> versions are in quay.io.<br>- The additional Operators to mirror. Select packages individually.<br>- The extra images that you want to add to the repository.</p><p>Example:</p><pre>apiVersion: mirror.openshift.io/v2alpha1<br>kind: ImageSetConfiguration<br>mirror:<br>  operators:<br>    - catalog: registry.redhat.io/redhat/redhat-operator-index:v4.18<br>      packages:<br>        - name: &quot;local-storage-operator&quot;<br>        - name: &quot;lvms-operator&quot;<br>        - name: &quot;kubernetes-nmstate-operator&quot;<br>        - name: &quot;redhat-oadp-operator&quot;<br>        - name: &quot;amq-streams&quot;<br>        - name: &quot;kubevirt-hyperconverged&quot;<br>        - name: &quot;metallb-operator&quot;<br>        - name: &quot;multicluster-engine&quot;<br>      full: true<br>  platform:<br>    channels:<br>    - name: stable-4.18<br>      minVersion: 4.18.10<br>      maxVersion: 4.18.10<br>      type: ocp<br>    kubeVirtContainer: true<br>    graph: true</pre><blockquote>Apply objects (IDMS/ICSP and Catalog Sources) in hub cluster.</blockquote><p>After the mirroring process is complete, you need to apply two objects in the management cluster:</p><p>- ImageContentSourcePolicy (ICSP) or ImageDigestMirrorSet (IDMS)<br>- Catalog sources</p><p>Once objects are applied in hub cluster, install operators given in the Prerequisites with corresponding custom resources.</p><blockquote>Configure TLS certification.</blockquote><p>To ensure proper function in a disconnected deployment, you need to configure the registry CA certificates in the management cluster</p><ol><li>Create a ConfigMap like given in following example:</li></ol><pre>kind: ConfigMap<br>apiVersion: v1<br>metadata:<br>  name: user-ca-bundle<br>  namespace: clusters<br>data:<br>  ca-bundle.crt: |<br>    // Registry1 CA<br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE----<br></pre><p>2. Patch the cluster-wide object, image.config.openshift.io to include <br> the following specification:</p><pre>spec:<br>  additionalTrustedCA:<br>    - name: registry-config</pre><blockquote>Create virtualized HCP cluster</blockquote><p>In OpenShift cluster deploy following resources.</p><p>a. Create Namespaces:<br> - one namespace will be clusters(default), <br> for example: clusters<br> - second namespace is clusters-hosted-cluster-name, <br> for example : clusters-disconnected1<br> Example:</p><pre>---<br>apiVersion: v1<br>kind: Namespace<br>metadata:<br>  creationTimestamp: null<br>  name: clusters-disconnected1<br>spec: {}<br>status: {}<br>---<br>apiVersion: v1<br>kind: Namespace<br>metadata:<br>  creationTimestamp: null<br>  name: clusters<br>spec: {}<br>status: {}</pre><p>b. Create a ConfigMap which contains registry CA certificates in hosted-cluster-namespace<br>Example:</p><pre>kind: ConfigMap<br>apiVersion: v1<br>metadata:<br>  name: user-ca-bundle<br>  namespace: clusters<br>data:<br>  ca-bundle.crt: |<br>    // Registry1 CA<br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE-----<br><br>    // Registry2 CA<br>    -----BEGIN CERTIFICATE-----<br>    -----END CERTIFICATE-----</pre><p>c. Create pull secret in hosted-cluster-namespace.<br> Example:</p><pre>kind: Secret<br>apiVersion: v1<br>metadata:<br>  name: disconnected-secret<br>  namespace: clusters<br>data:<br>  .dockerconfigjson: xxxxxxxx<br>type: kubernetes.io/dockerconfigjson</pre><p>d. Create HostedCluster CR by replacing appropriate values.<br>Example:</p><pre>apiVersion: hypershift.openshift.io/v1beta1<br>kind: HostedCluster<br>metadata:<br>  name: disconnected1<br>  namespace: clusters<br>spec:<br>  fips: false<br>  release:<br>    image: $TARGET_PATH/openshift/release-images:&lt;ocp_version&gt;-x86_64<br>  dns:<br>    baseDomain: &lt;xyz.com&gt;<br>  controllerAvailabilityPolicy: HighlyAvailable<br>  etcd:<br>    managed:<br>      storage:<br>        persistentVolume:<br>          size: 8Gi<br>          storageClassName: ocs-storagecluster-ceph-rbd<br>        type: PersistentVolume<br>    managementType: Managed<br>  infrastructureAvailabilityPolicy: SingleReplica<br>  platform:<br>    kubevirt:<br>      baseDomainPassthrough: true<br>    type: KubeVirt<br>  additionalTrustBundle:<br>    name: user-ca-bundle<br>  networking:<br>    clusterNetwork:<br>      - cidr: 10.132.0.0/14<br>    networkType: OVNKubernetes<br>    serviceNetwork:<br>      - cidr: 172.31.0.0/16<br>  pullSecret:<br>    name: disconnected-secret<br>  configuration:<br>    operatorhub:<br>      disableAllDefaultSources: true<br>  capabilities: {}<br>  sshKey:<br>    name: &#39;&#39;<br>  autoscaling: {}<br>  imageContentSources:<br>    - mirrors:<br>        - $TARGET_PATH/openshift/release<br>      source: quay.io/openshift-release-dev/ocp-v4.0-art-dev<br>    - mirrors:<br>        - $TARGET_PATH/redhat<br>      source: registry.redhat.io/redhat<br>    - mirrors:<br>        - $TARGET_PATH/rhel9<br>      source: registry.redhat.io/rhel9<br>    - mirrors:<br>        - $TARGET_PATH/rhel8<br>      source: registry.redhat.io/rhel8<br>    - mirrors:<br>        - $TARGET_PATH/openshift/release-images<br>      source: quay.io/openshift-release-dev/ocp-release<br>    - mirrors:<br>        - $TARGET_PATH/openshift4<br>      source: registry.redhat.io/openshift4<br>    - mirrors:<br>        - $TARGET_PATH/multicluster-engine<br>      source: registry.redhat.io/multicluster-engine<br>  olmCatalogPlacement: management<br>  services:<br>    - service: APIServer<br>      servicePublishingStrategy:<br>        type: LoadBalancer<br>    - service: Ignition<br>      servicePublishingStrategy:<br>        type: Route<br>    - service: Konnectivity<br>      servicePublishingStrategy:<br>        type: Route<br>    - service: OAuthServer<br>      servicePublishingStrategy:<br>        type: Route</pre><p>Note: While creating HostedCluster CR, please remove OIDC service.</p><p>e. Create Nodepool CR by replacing.</p><pre>apiVersion: hypershift.openshift.io/v1beta1<br>kind: NodePool<br>metadata:<br> name: disconnected1<br> namespace: clusters<br>spec:<br> arch: amd64<br> clusterName: disconnected1<br> management:<br>   autoRepair: false<br>   upgradeType: Replace<br> nodeDrainTimeout: 0s<br> nodeVolumeDetachTimeout: 0s<br> platform:<br>   kubevirt:<br>     attachDefaultNetwork: true<br>     compute:<br>       cores: 2<br>       memory: 6Gi<br>     networkInterfaceMultiqueue: Enable<br>     rootVolume:<br>       persistent:<br>         size: 32Gi<br>       type: Persistent<br>   type: KubeVirt<br> release:<br>   image: $TARGET_PATH/openshift/release-images:&lt;ocp_version&gt;-x86_64<br> replicas: 2</pre><p>After 20–30 mins virtualized HCP cluster will be created.</p><p>That’s how we simply enable secure, disconnected OpenShift clusters using virtualized Hosted Control Planes on IBM Fusion HCI. By sharing infrastructure without sacrificing isolation, teams can scale faster, operate independently, and meet even the strictest security requirements with ease.</p><p>Interested in how data services are managed on hosted control planes in IBM Fusion’s disconnected environments? <br>Follow along to find out more: <a href="https://medium.com/@spp17081978/managing-data-services-on-hosted-control-planes-in-ibm-fusion-in-disconnected-environment-9d837ccfab1b">https://medium.com/@spp17081978/managing-data-services-on-hosted-control-planes-in-ibm-fusion-in-disconnected-environment-9d837ccfab1b</a></p><p>Resources: <br><a href="https://www.ibm.com/docs/en/fusion-hci-systems/2.12.0?topic=mcfhuhcp-deploying-virtualized-hosted-control-plane-cluster-in-disconnected-environment">https://www.ibm.com/docs/en/fusion-hci-systems/2.12.0?topic=mcfhuhcp-deploying-virtualized-hosted-control-plane-cluster-in-disconnected-environment</a></p><p>Thank you for taking the time to read this blog. I appreciate your interest and hope you found the information helpful and easy to follow. Your engagement is invaluable, and I encourage you to reach out with any questions or feedback. Your input helps improve future content and ensures we continue to provide useful and relevant information.</p><p>𝐃𝐢𝐬𝐜𝐥𝐚𝐢𝐦𝐞𝐫: The views and opinions expressed in this guide are solely my own and do not reflect those of IBM. This is shared for informational purposes only and should not be considered as official guidance or advice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6b04b522a3ba" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Create a Bare Metal Hosted Control Plane using DELL hardware within the IBM Fusion HCI]]></title>
            <link>https://medium.com/@siddhic24/create-bare-metal-hosted-control-plane-with-dell-hardware-in-ibm-fusion-hci-c4e8be4f5618?source=rss-f77915d403e9------2</link>
            <guid isPermaLink="false">https://medium.com/p/c4e8be4f5618</guid>
            <dc:creator><![CDATA[Siddhi Chavan]]></dc:creator>
            <pubDate>Sun, 06 Jul 2025 13:45:24 GMT</pubDate>
            <atom:updated>2025-08-03T03:42:48.174Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/882/1*TOBSIhy7Hb_Dt6TMCzXhzQ.png" /></figure><blockquote><strong>Introduction</strong></blockquote><p>As enterprises push for scalable, high-performance infrastructure, the need for versatile deployment options continues to grow. IBM Fusion HCI is an all-in-one appliance for deploying OpenShift addresses this demand by bundling everything needed for an enterprise-grade OpenShift cluster: compute, networking, storage, GPU, and data resiliency. Its Hosted Control Plane (HCP) capability allows organizations to carve the appliance into multiple OpenShift clusters, offering flexibility and efficiency at scale.</p><p>Now, Fusion HCI is raising the bar by supporting Bare Metal HCP on Dell nodes. This evolution brings increased freedom of choice to infrastructure architects: run Hosted Control Planes on virtualized platforms or directly on Dell physical hardware — whichever suits your workload and performance profile best.</p><p>In this blog, we’ll walk through the process of setting up a bare metal HCP environment using Dell infrastructure, showcasing how IBM Fusion HCI makes it faster, easier, and more resilient than ever before.</p><blockquote><strong>Key steps covered</strong></blockquote><ol><li>Installation of Prerequisites</li><li>Network Configuration</li><li>Host Inventory Configuration</li><li>Creating the Infrastructure Environment</li><li>NMStateConfig Creation</li><li>Adding Hosts to Inventory</li><li>Creating the Bare Metal HCP Cluster</li><li>Ingress Configuration</li></ol><blockquote><strong>Prerequisites</strong></blockquote><ol><li>multicluster engine for Kubernetes (MCE) Operator with it’s multiclusterengine CR.</li><li>MetalLB operator with its 3 CRs: metallb, IPAddressPool and L2Advertisement.</li></ol><blockquote><strong>Network Settings</strong></blockquote><p>Before starting the deployment, make sure to create DNS entries for:</p><ul><li>Host that will be used as a node in the Bare Metal HCP cluster deployment.</li><li>Ingress. (This IP address is used to create a load balancer on the Hosted Control Plane cluster to allow ingress to that cluster.)</li></ul><blockquote><strong>Configuring Host Inventory</strong></blockquote><p>To configure host inventory, follow below steps:</p><ul><li>Go to MCE (All Clusters) view, and from menu click on host inventory.</li><li>In the host inventory window, there will be an option as configure host inventory settings.</li><li>Put values for Database, System, Image storage and click on configure.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SRyae6zkVqdn7Ktv_d3N6g.png" /><figcaption>Configure host inventory settings</figcaption></figure><blockquote><strong>Create Infrastructure Environment</strong></blockquote><ul><li>Once host inventory is configured, next step is to create infrastructure environment.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r-_jdhOT6QNqeatWuPNu9Q.png" /><figcaption>Create infrastructure environment</figcaption></figure><ul><li>After this, enter the following details and click on create:<br><strong>Name</strong>: Name of the infrastructure.<br><strong> Network type</strong>: Select <strong>Static IP</strong>, <strong>Bridges</strong>, and <strong>Bonds</strong>.<br><strong>Location</strong>: location according to the user.<br><strong>Label</strong>: Optional value.<br><strong>Pull secret</strong>: <br> The pull secret must include the following:<br> 1. cloud.openshift.com<br> 2. cp.icr.io<br> 3. quay.io<br> 4. registry.connect.redhat.com<br> 5. registry.redhat.io<br><strong>SSH Public key</strong>: Generate a private public ssh key pair from bastion host and specify public key in ssh field. It is optional but recommended.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Lsy_Q1uK-C8GgvU0-bkU1w.png" /><figcaption>Host inventory</figcaption></figure><blockquote><strong>Create NMStateConfig for each host</strong></blockquote><p>The NMStateConfig allows the server to be recognized by the <br> IBM Fusion HCI cluster.</p><p><strong>NOTE:</strong> The NMStateConfig varies based on whether static IP assignment or DHCP IP assignment is used.</p><p>Example of static IP NMStateConfig:</p><pre>apiVersion: agent-install.openshift.io/v1beta1<br>kind: NMStateConfig<br>metadata:<br>  labels:<br>    infraenvs.agent-install.openshift.io: &lt;same as infra name&gt;<br>  name: &lt;any unique name for host within infra for example fusion-bm-ru8&gt;<br>  namespace: &lt;same namespace as used for infra&gt;<br>spec:<br>  config:<br>    dns-resolver:<br>      config:<br>        server:<br>          - &lt;DNS server IP&gt;<br>    interfaces:<br>      - ipv6:<br>          enabled: false<br>        link-aggregation:<br>          mode: 802.3ad<br>          options:<br>            lacp_rate: &#39;1&#39;<br>            miimon: &#39;140&#39;<br>            xmit_hash_policy: &#39;1&#39;<br>          ports:<br>            - &lt;Bare metal Primary interface’s first port/interface name&gt;<br>            - &lt;Bare metal Primary interface’s second port/interface name&gt;<br>        mac-address: &lt;Bare metal Primary interface’s first port’s Mac address&gt;<br>        mtu: &lt;MTU value&gt;<br>        name: bond0<br>        state: up<br>        type: bond<br>      - ipv4:<br>          address:<br>            - ip: &lt;Bare metal primary interface IP&gt;<br>              prefix-length: 24<br>          dhcp: false<br>          enabled: true<br>        ipv6:<br>          enabled: false<br>        name: &lt;bond name followed by VLAN ID of type OpenShift Customer VLAN present on the cluster for example bond0.VLAN_ID&gt; <br>        state: up<br>        type: vlan<br>        vlan:<br>          base-iface: bond0<br>          id: &lt;VLAN ID of type OpenShift Customer VLAN present on the cluster&gt;<br>    routes:<br>      config:<br>        - destination: 0.0.0.0/0<br>          next-hop-address: &lt;Gateway IP&gt;<br>          next-hop-interface: &lt;bond name followed by VLAN ID of type OpenShift Customer VLAN present on the cluster for example bond0.VLAN_ID&gt;<br>  interfaces:<br>    - macAddress: &lt;Bare metal Primary interface’s first port’s Mac address&gt;<br>      name: Bare metal Primary interface’s first port/interface name&gt;<br>    - macAddress: &lt;Bare metal Primary interface’s second port’s Mac address&gt;<br>      name: Bare metal Primary interface’s second port/interface name&gt;</pre><blockquote><strong>Add host to the host inventory</strong></blockquote><ul><li>After creation of Infrastructure environment and application of NMStateConfig for each host, add host in the host inventory.</li><li>To add host, click on add host with discovery ISO. Download the full ISO.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mbtfWGeR6nmLeqYWjpdYkQ.png" /><figcaption>Add host with Discovery ISO</figcaption></figure><ul><li>Boot the host with downloaded discovery ISO.<br>To do this step, log in to iDRAC -&gt; go to remote console -&gt; mount ISO image -&gt; reboot the server.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_Lg8MnpvfUaToJXHQZWM2A.png" /><figcaption>After rebooting server</figcaption></figure><ul><li>Once reboot of a host is completed, host will be visible in the host inventory. After this approve host to make it available to use for deployment of Bare Metal HCP cluster.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2UuCo7wZwq7DUfgUfPiRHg.png" /><figcaption>Approve host</figcaption></figure><ul><li>Follow same procedure mentioned in “<strong>Create NMStateConfig for each host” </strong>and<strong> “Add host to the host inventory” </strong>for each host.</li></ul><blockquote><strong>Create Bare Metal HCP cluster:</strong></blockquote><p>Now that hosts are available, next step is to create Bare Metal HCP cluster using following steps:</p><ul><li>Go to clusters -&gt; create cluster -&gt; host inventory -&gt; hosted -&gt; fill necessary details:</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Tw2qoWZrDb-gtV2X6TROtw.png" /><figcaption>Cluster Details</figcaption></figure><ul><li>Next enter Node pool details:</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7-3m6C1gKytU7lWKn6Ri0w.png" /><figcaption>Node pool details</figcaption></figure><ul><li>Next enter Networking details:</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rbq3uq0GjRAdbrJR-Y32Hg.png" /><figcaption>Network details</figcaption></figure><ul><li>Then simply review and create cluster. After some time cluster will be created.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n2jTnO9yUS3W4wuHNZvC_g.png" /><figcaption>BM HCP Cluster is in ready state</figcaption></figure><blockquote><strong>Configuring ingress for Bare Metal HCP cluster</strong></blockquote><p>After deployment of BM HCP cluster, you must configure the ingress for you to view the changes in the user interface.</p><p>To configure ingress follow below steps:<br>1. Log in into HCP cluster through cluster API using command:</p><pre>oc login --server=&lt;cluster_api_address&gt; -u &lt;username&gt; -p &lt;password&gt;</pre><p>2. Create MetalLB operator using following steps.<br> a. Create a new YAML file to deploy MetalLB operator. <br> Example YAML:</p><pre>apiVersion: v1<br>kind: Namespace<br>metadata:<br>  name: metallb<br>  labels:<br>    openshift.io/cluster-monitoring: &quot;true&quot;<br>  annotations:<br>    workload.openshift.io/allowed: management<br>---<br>apiVersion: operators.coreos.com/v1<br>kind: OperatorGroup<br>metadata:<br>  name: metallb-operator-operatorgroup<br>  namespace: metallb<br>---<br>apiVersion: operators.coreos.com/v1alpha1<br>kind: Subscription<br>metadata:<br>  name: metallb-operator<br>  namespace: metallb<br>spec:<br>  channel: &quot;stable&quot;<br>  name: metallb-operator<br>  source: redhat-operators<br>  sourceNamespace: openshift-marketplace</pre><p>b. Apply the YAML using command:</p><pre>oc apply -f metallb-operator-config.yaml</pre><p>c. Wait for all the pods to be up and running. Run the following command to confirm:</p><pre>oc get pods -n metallb</pre><p>3. Create an instance of metalLB using following steps:<br> a. Create a new file metallb-instance-config.yaml. <br> Example:</p><pre><br>apiVersion: metallb.io/v1beta1<br>kind:  MetalLB<br>metadata:<br>   name: metallb<br>   namespace:  metallb</pre><p>b. Apply the file using following command:</p><pre>oc apply -f metallb-instance-config.yaml</pre><p>c. Wait for the pods to be up and running.</p><pre>oc get pods -n metallb</pre><p>4. Create an IPAddressPool and L2Advertisementinstances using following steps:<br> a. Create an IPAddressPool named ipaddresspool-l2advertisement-config.yaml. <br> Example:</p><pre>apiVersion: metallb.io/v1beta1<br>kind: IPAddressPool<br>metadata:<br>  name: ippool<br>  namespace: metallb<br>spec:<br>  protocol: layer2<br>  autoAssign: false<br>  addresses:<br>    - 1.23.45.678-1.23.45.678<br>---<br>apiVersion: metallb.io/v1beta1<br>kind: L2Advertisement<br>metadata:<br>  name: l2-advertisement<br>  namespace: metallb<br>spec:<br>  ipAddressPools:<br>    - ippool</pre><p><strong>NOTE: The name of the </strong><strong>IPAddressPool must match the one in the </strong><strong>L2Advertisement. The addresses must be the IPAddress in the DNS that is the </strong><strong>*.apps.NAMEofCluster in the DNS network table set up by the network administrator.</strong></p><p>b. Apply the ipaddresspool-l2advertisement-config.yaml.</p><pre>oc apply -f ipaddresspool-l2advertisement-config.yaml</pre><p>5. Create a service for the loadbalancerusing following steps:<br> a. Create metallb-loadbalancer-service.yaml.</p><pre>kind: Service<br>apiVersion: v1<br>metadata:<br>  annotations:<br>    metallb.universe.tf/address-pool: ippool<br>  name: metallb-ingress<br>  namespace: openshift-ingress<br>spec:<br>  ports:<br>    - name: http<br>      protocol: TCP<br>      port: 80<br>      targetPort: 80<br>    - name: https<br>      protocol: TCP<br>      port: 443<br>      targetPort: 443<br>  selector:<br>    ingresscontroller.operator.openshift.io/deployment-ingresscontroller: default<br>  type: LoadBalancer</pre><p><strong>NOTE: The </strong><strong>address-pool must match the name of the </strong><strong>addresspool in previous step.</strong></p><p>b. Apply the load balancer service YAML.</p><pre>oc apply -f metallb-loadbalancer-service.yaml</pre><p>c. Monitor the Cluster Operator console for issues:</p><pre>oc get co</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pnvvKJHhJHYNu-6jlL_mlg.png" /><figcaption>Ingress is available</figcaption></figure><p>After Ingress and console are available, Login to Bare Metal HCP cluster using username and password.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WdYtPRUwPXUkQhDGRfNygA.png" /><figcaption>BM HCP Cluster</figcaption></figure><blockquote>References:</blockquote><p><a href="https://www.ibm.com/docs/en/fusion-hci-systems/2.10.0?topic=plane-deployment-bare-metal-clusters-fusion-data-foundation">https://www.ibm.com/docs/en/fusion-hci-systems/2.10.0?topic=plane-deployment-bare-metal-clusters-fusion-data-foundation</a></p><p>Thank you for taking the time to read this blog. I appreciate your interest and hope you found the information helpful and easy to follow. Your engagement is invaluable, and I encourage you to reach out with any questions or feedback. Your input helps improve future content and ensures we continue to provide useful and relevant information.</p><p>𝐃𝐢𝐬𝐜𝐥𝐚𝐢𝐦𝐞𝐫: The views and opinions expressed in this guide are solely my own and do not reflect those of IBM. This is shared for informational purposes only and should not be considered as official guidance or advice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c4e8be4f5618" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Configuring PCI Passthrough to allow Virtual Machines (VMs) to access a GPU in IBM Fusion HCI]]></title>
            <link>https://medium.com/@siddhic24/configuring-pci-passthrough-to-allow-virtual-machines-vms-to-access-a-gpu-in-ibm-fusion-hci-eced7500d215?source=rss-f77915d403e9------2</link>
            <guid isPermaLink="false">https://medium.com/p/eced7500d215</guid>
            <dc:creator><![CDATA[Siddhi Chavan]]></dc:creator>
            <pubDate>Tue, 01 Apr 2025 11:29:27 GMT</pubDate>
            <atom:updated>2025-04-01T12:51:39.468Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/882/1*TOBSIhy7Hb_Dt6TMCzXhzQ.png" /></figure><h3><strong>Introduction:</strong></h3><blockquote>IBM Fusion HCI is a flexible platform that simplifies managing virtualized workloads, supporting both containers and virtual machines efficiently. It integrates seamlessly with OpenShift Virtualization (OSV) to scale and optimize resources, making it ideal for large-scale inferencing workloads, while also providing easy access to GPU acceleration.</blockquote><blockquote>When GPUs are enabled in virtual machines, they significantly boost the computational power available to those VMs, supporting demanding workloads.</blockquote><blockquote>When using GPUs in virtual machines, two common methods are PCI Passthrough and Virtual GPUs (vGPU). PCI Passthrough allows a virtual machine to access a physical GPU directly, providing it with near-native performance. This method ensures that the virtual machine has exclusive access to the GPU, maximizing performance for resource-intensive applications.</blockquote><blockquote>This blog will primarily focus on the process of configuring PCI Passthrough within virtual machines to enable GPU.</blockquote><h3><strong>What you’ll learn:</strong></h3><p>In this guide, I will walk through the process of enabling PCI Passthrough inside virtual machines to ensure GPU access. The approach includes:</p><ol><li>Installation of Prerequisites.</li><li>Preparing host devices for PCI passthrough.</li><li>Exposing PCI host devices in the cluster.</li><li>Configuring virtual machines for PCI passthrough.</li></ol><h3><strong>Step-by-step setup: configuring PCI Passthrough within virtual machines to access GPUs.</strong></h3><blockquote><strong>Step 1: Install Prerequisites</strong></blockquote><ul><li><strong>OpenShift cluster with GPU node.</strong></li><li><strong>NVIDIA GPU Operator: </strong>(Note: Do not create any default cluster policy)<br>The NVIDIA GPU Operator automates the deployment of GPU drivers and tools in Kubernetes, enabling seamless GPU access for AI and machine learning workloads.</li><li><strong>Node Feature Discovery Operator by RedHat and it’s default CR:<br></strong>The Node Feature Discovery (NFD) Operator detects and exposes hardware features of the nodes, helping optimize workload placement based on node capabilities like CPU and GPU availability.</li><li><strong>OpenShift Virtualization Operator and it’s HyperConverged CR:</strong><br>OpenShift Virtualization enables running and managing virtual machines alongside containers in OpenShift. It integrates traditional and containerized workloads, streamlining VM lifecycle management and resource optimization.</li></ul><blockquote><strong>Step 2: Preparing host devices for PCI passthrough</strong></blockquote><blockquote><strong>2.1: Create and apply MachineConfig which enables IOMMU.</strong></blockquote><ul><li>Create a MachineConfig for example 100-worker-iommu.yaml to add kernel arguments to enable the Input-Output Memory Management Unit (IOMMU).</li></ul><p><em>Note: kernel Argument value will depend on processor type. If It is intel then value is supposed to be intel_iommu=on However if it is amd processor then it would be amd_iommu=on.</em></p><p>Example file:</p><pre>apiVersion: machineconfiguration.openshift.io/v1<br>kind: MachineConfig<br>metadata:<br>  labels:<br>    machineconfiguration.openshift.io/role: worker <br>  name: 100-worker-iommu <br>spec:<br>  config:<br>    ignition:<br>      version: 3.2.0<br>  kernelArguments:<br>      - intel_iommu=on</pre><ul><li>Apply the MachineConfig 100-worker-iommu.yaml created using <br>command:</li></ul><pre>oc apply -f 100-worker-iommu.yaml</pre><ul><li>Verify if kernel argument is enabled using command:</li></ul><pre>cat /proc/cmdline</pre><p>Example output: intel_iommu=on</p><blockquote><strong>2.2: Bind the PCI devices to the Virtual Function I/O (VFIO) driver.</strong></blockquote><ul><li>To bind PCI devices to the VFIO (Virtual Function I/O) driver, obtain the value of Device-ID using command:</li></ul><pre>lspci -nnkd ::0302</pre><p>Example output: Here device-id is <strong>10de:20b5</strong></p><pre>3d:00.0 3D controller [0302]: NVIDIA Corporation GA100 [A100 PCIe 80GB] [10de:20b5] (rev a1)<br>	Subsystem: NVIDIA Corporation Device [10de:1533]<br>63:00.0 3D controller [0302]: NVIDIA Corporation GA100 [A100 PCIe 80GB] [10de:20b5] (rev a1)<br>	Subsystem: NVIDIA Corporation Device [10de:1533]<br>ab:00.0 3D controller [0302]: NVIDIA Corporation GA100 [A100 PCIe 80GB] [10de:20b5] (rev a1)<br>	Subsystem: NVIDIA Corporation Device [10de:1533]</pre><blockquote><strong>2.3: Bind the PCI devices to the VFIO driver.</strong></blockquote><ul><li>Create a Butane config file for example 100-worker-vfiopci.bu for binding the PCI devices to the VFIO driver.</li></ul><pre>variant: openshift<br>version: 4.16.0<br>metadata:<br>  name: 100-worker-vfiopci<br>  labels:<br>    machineconfiguration.openshift.io/role: worker <br>storage:<br>  files:<br>  - path: /etc/modprobe.d/vfio.conf<br>    mode: 0644<br>    overwrite: true<br>    contents:<br>      inline: |<br>        options vfio-pci ids=10de:20b5<br>  - path: /etc/modules-load.d/vfio-pci.conf <br>    mode: 0644<br>    overwrite: true<br>    contents:<br>      inline: vfio-pci</pre><ul><li>Use Butane to generate a MachineConfig file for example 100-worker-vfiopci.yaml using command:</li></ul><pre>butane 100-worker-vfiopci.bu -o 100-worker-vfiopci.yaml</pre><ul><li>Apply the MachineConfig to the worker nodes using command:</li></ul><pre>oc apply -f 100-worker-vfiopci.yaml</pre><ul><li>Verify if VFIO driver in loaded using command:</li></ul><pre>lspci -nnkd ::0302</pre><p>Example output: Kernel driver in use: vfio-pci</p><pre>3d:00.0 3D controller [0302]: NVIDIA Corporation GA100 [A100 PCIe 80GB] [10de:20b5] (rev a1)<br>	Subsystem: NVIDIA Corporation Device [10de:1533]<br>	Kernel driver in use: vfio-pci<br>	Kernel modules: nouveau<br>63:00.0 3D controller [0302]: NVIDIA Corporation GA100 [A100 PCIe 80GB] [10de:20b5] (rev a1)<br>	Subsystem: NVIDIA Corporation Device [10de:1533]<br>	Kernel driver in use: vfio-pci<br>	Kernel modules: nouveau<br>ab:00.0 3D controller [0302]: NVIDIA Corporation GA100 [A100 PCIe 80GB] [10de:20b5] (rev a1)<br>	Subsystem: NVIDIA Corporation Device [10de:1533]<br>	Kernel driver in use: vfio-pci<br>	Kernel modules: nouveau</pre><blockquote><strong>Step 3: Exposing PCI host devices in the cluster</strong></blockquote><ul><li>Add PCI device information to the spec.permittedHostDevices.pciHostDevices array of HyperConverged CR and save changes.</li></ul><p>Example configuration file:</p><pre>apiVersion: hco.kubevirt.io/v1<br>kind: HyperConverged<br>metadata:<br>  name: kubevirt-hyperconverged<br>  namespace: openshift-cnv<br>spec:<br>  permittedHostDevices:<br>    pciHostDevices:<br>    - pciDeviceSelector: &#39;10de:20b5&#39;<br>      resourceName: nvidia/a100<br># ...</pre><blockquote><strong>Step 4: Configuring virtual machines for PCI passthrough</strong></blockquote><ul><li>Create VM in OpenShift Virtualization and add GPU device by clicking customize VM button.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3ENoWdgMDyn0EVk1vE0B-Q.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TXAEEuanXe2oZ7qpRmwFcw.png" /></figure><ul><li>Go to VM’s terminal to verify if GPU device is accessible inside VM using command:</li></ul><pre>lspci -nnkd ::0302</pre><p>Example outcome:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dH7EZmJQGFAgneoYfePp1A.png" /></figure><p>Thank you for taking the time to read this blog. I appreciate your interest and hope you found the information helpful and easy to follow. Your engagement is invaluable, and I encourage you to reach out with any questions or feedback. Your input helps improve future content and ensures we continue to provide useful and relevant information.</p><p>𝐃𝐢𝐬𝐜𝐥𝐚𝐢𝐦𝐞𝐫: The views and opinions expressed in this guide are solely my own and do not reflect those of my employer. This is shared for informational purposes only and should not be considered as official guidance or advice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eced7500d215" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Harnessing the Power of OpenShift Virtualization on IBM Fusion: A Step-by-Step Guide to Deploying…]]></title>
            <link>https://medium.com/@siddhic24/harnessing-the-power-of-openshift-virtualization-on-ibm-fusion-a-step-by-step-guide-to-deploying-9c914a341abd?source=rss-f77915d403e9------2</link>
            <guid isPermaLink="false">https://medium.com/p/9c914a341abd</guid>
            <dc:creator><![CDATA[Siddhi Chavan]]></dc:creator>
            <pubDate>Tue, 30 Jul 2024 08:54:39 GMT</pubDate>
            <atom:updated>2024-08-29T07:18:11.271Z</atom:updated>
            <content:encoded><![CDATA[<h3>Harnessing the Power of OpenShift Virtualization on IBM Fusion: A Step-by-Step Guide to Deploying Real-World Workloads (LAMP)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*shzKYBdySRca5zDsQ_JTug.png" /></figure><h3><strong>Introduction:</strong></h3><blockquote>This blog provides a step-by-step guide on creating a virtual machine and a workload. We’ll learn how to create virtual machine with a workload which will be running on the top of it.</blockquote><h3><strong>About Demo workload:</strong></h3><blockquote>The LAMP stack is a set of tools used to create and manage websites and web applications. It includes Linux (the operating system), Apache (the web server that handles web requests), MySQL or MariaDB (the database that stores information), and PHP (the language that makes websites interactive). Together, these components provide a reliable and flexible foundation for building websites.</blockquote><blockquote>We will showcase this sample workload using the LAMP stack for demonstration purposes. This setup will include the installation of HTTPS, MariaDB and PHP. After successfully deploying the virtual machine and running the workload, the output will provide information about PHP.</blockquote><h3><strong>Creating a virtual machine with workload:</strong></h3><ul><li>Prepare a YAML file for creating a virtual machine. Here we are taking a example yaml of RHEL 9.</li></ul><pre>apiVersion: kubevirt.io/v1<br>kind: VirtualMachine<br>metadata:<br>  labels:<br>    app: demo-rhel   #&lt;vm-name&gt;<br>  name: demo-rhel   #&lt;vm-name&gt;<br>spec:<br>  dataVolumeTemplates:<br>  - apiVersion: cdi.kubevirt.io/v1beta1<br>    kind: DataVolume<br>    metadata:<br>      name: demo-rhel   #&lt;vm-name&gt;<br>    spec:<br>      sourceRef:<br>        kind: DataSource<br>        name: rhel9<br>        namespace: openshift-virtualization-os-images<br>      storage:<br>        resources:<br>          requests:<br>            storage: 30Gi<br>  running: true  <br>  template:<br>    metadata:<br>      labels:<br>        kubevirt.io/domain: example-2<br>    spec:<br>      domain:<br>        cpu:<br>          cores: 1<br>          sockets: 2<br>          threads: 1<br>        devices:<br>          disks:<br>          - disk:<br>              bus: virtio<br>            name: rootdisk<br>          - disk:<br>              bus: virtio<br>            name: cloudinitdisk<br>          interfaces:<br>          - masquerade: {}<br>            name: default<br>          rng: {}<br>        features:<br>          smm:<br>            enabled: true<br>        firmware:<br>          bootloader:<br>            efi: {}<br>        resources:<br>          requests:<br>            memory: 8Gi<br>      evictionStrategy: LiveMigrate<br>      networks:<br>      - name: default<br>        pod: {}<br>      volumes:<br>      - dataVolume:<br>          name: demo-rhel   #&lt;vm-name&gt;<br>        name: rootdisk<br>      - cloudInitConfigDrive:<br>            secretRef:<br>              name: demo-secret  #&lt;secret filename&gt;<br>        name: cloudinitdisk</pre><p><strong>Question:</strong> This is a yaml file for creation of vitual machine in openshift virtualization but can we deploy a workload in the same YAML ? How ?</p><p>Answer is <strong>YES.</strong></p><p><strong>Workloads can be deployed with the same YAML using following steps:</strong></p><ul><li>Create a bash script which will contain commands to deploy a workload. Below providing one example bash script which we used to run a sample workload.</li></ul><pre>#!/bin/bash<br>export NEW_USER=&quot;cloud-user&quot;<br>sudo adduser -U -m $NEW_USER<br>sudo echo &quot;$NEW_USER:`echo Y2xvdWQtdXNlcg== | base64 -d`&quot; | chpasswd<br><br>sudo subscription-manager register --user=&lt;username&gt; --password=`echo &lt;password&gt; | base64 -d` --serverurl=https://subscription.rhsm.redhat.com --force --auto-attach &gt;&gt; /tmp/subscription_log.out<br><br>sudo bash<br>sudo yum update -y<br>sudo yum install httpd -y <br>sudo yum install mariadb-server -y<br>sudo yum install php -y<br>sudo yum install php-mysqlnd -y<br><br>sudo systemctl start mariadb<br>sudo systemctl enable mariadb<br>sudo systemctl start httpd.service<br>sudo systemctl enable httpd.service<br><br>sudo mysql -e &quot;CREATE DATABASE mydb; CREATE USER &#39;myuser&#39;@&#39;localhost&#39; IDENTIFIED BY &#39;mypassword&#39;; GRANT ALL PRIVILEGES ON mydb.* TO &#39;myuser&#39;@&#39;localhost&#39;; FLUSH PRIVILEGES;&quot;<br>sudo echo &#39;&lt;?php phpinfo(); ?&gt;&#39; &gt; /var/www/html/info.php</pre><ul><li>Go to OpenShift cluster → workloads → secrets.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bOkzvdJ659kmj-MSmRV6MQ.png" /></figure><ul><li>Select project (For example: default as per the VM namespace) and click create button which is placed on the right top corner.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9_03OCeewm4FNRxO3ZvTqA.png" /></figure><ul><li>Select “key/value“ secret and fill all the fields and place bash script and save it.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0YC4EYnImF3C1ZBT8eASFw.png" /></figure><ul><li>If we see above yaml file for virtual machine creation then at the bottom we can look at following lines. <strong>This is where we have to add our secret file in order to deploy workload within creation yaml file.</strong></li></ul><pre>- cloudInitConfigDrive:<br>            secretRef:<br>              name: demo-secret  #&lt;secret filename&gt;<br>          name: cloudinitdisk</pre><ul><li>Now our virtual machine is ready to deploy.</li><li>Now go to virtualization → VirtualMachines</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OovJ03y376vYZQSqJzBmzQ.png" /></figure><ul><li>Click create which is placed at right top corner in order to create the virtual machine. In that select <strong>“With YAML“</strong> option.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8eIGeY-K9XVI9wydWX7Xww.png" /></figure><ul><li>Paste yaml created by you and hit create. That is how the virtual machine is created with workload running on top of it.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CHH8PhhkbwAWd7eEbMZFTA.png" /></figure><ul><li>Check status of the virtual machine.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0G4viqWXIsWJkAGDGYIXJg.png" /></figure><ul><li>Click on the virtual machine name for example <strong>“demo-rhel”</strong> to see the complete view of the virtual machine.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gsygVwEiJ2TK6_cTkjr3MA.png" /></figure><h3><strong>Verify if workload is deployed successfully :</strong></h3><ul><li>To verify that a virtual machine is created with a running workload, we need to execute certain commands to check the outcome.</li><li>There are two ways to check the outcome:</li></ul><p><strong>A. Calling endpoint from another pod</strong></p><ul><li>Step 1: Create one sample pod. For that go to workloads → pods. Create one pod. For example we have created a nginx pod named <strong>“example-test”</strong> to run a workload.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Y3eV0oEJSQau2rBDV2F0MQ.png" /></figure><ul><li>Step 2: Take IPAddress of the virtual machine from UI or by using command “<strong>oc get vmi -n &lt;namespace&gt;”. </strong>Here one can find the VM name with IPAddress.</li><li>Step 3: Run following command on the pod’s terminal.</li></ul><pre>curl &lt;VM&#39;S IPADRESS&gt;/info.php</pre><ul><li>Result: will print information about PHP after calling phpinfo().</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bnMd5E4peKfYfM7IZlOJEw.png" /></figure><p><strong>B. With the help of command line.</strong></p><ul><li>Step 1: Go to the local terminal and log into the HCI cluster with the help of token. To get token go to the cluster UI → kubeadmin (placed at right top corner) → copy login command.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JzyHNDZbmuvs7rQm4uiEQg.png" /></figure><ul><li>Step 2: Simply paste the login command in the local terminal.</li><li>Step 3: Log into your virtual machine with the help of following command:</li></ul><pre>virtctl console &lt;VM_NAME&gt; -n &lt;namespace&gt;</pre><ul><li>Step 3: After this run following command to see the outcome</li></ul><pre>curl localhost/info.php</pre><ul><li>Result:</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UCe1WfL4BUXolfpit2K4_A.png" /></figure><h3>Troubleshooting:</h3><p>If anyone encounters any error while running workload on CLI then try out running commands from bash script in order to verify if everything is up and running.</p><p>For example, if we find “curl localhost/info.php” command is not working then I should see first if apache, mariadb, mysql and php is installed properly or not and they are in running state or not by running following commands.</p><pre>sudo systemctl start mariadb<br>sudo systemctl enable mariadb<br>sudo systemctl start httpd.service<br>sudo systemctl enable httpd.service</pre><p>Thank you for taking the time to read this blog and explore the process of setting up of virtual machines and deploying workloads in OpenShift Virtualization. I appreciate your interest and hope you found the information helpful and easy to follow.</p><p>Your engagement is invaluable, and I encourage you to reach out with any questions or feedback. Your input helps improve future content and ensures we continue to provide useful and relevant information.</p><p>Special thanks to <a href="https://medium.com/u/4e4d6f550305"><strong>Sandeep Bazar</strong></a> and <a href="https://medium.com/u/b4819afdd1f0"><strong>ROHAN SHINDE</strong></a> for their invaluable contributions. Their expertise has significantly enhanced the accuracy and depth of this concept.</p><p><strong>Thank you again for all the support.</strong></p><p>𝐃𝐢𝐬𝐜𝐥𝐚𝐢𝐦𝐞𝐫: The views and opinions expressed in this guide are solely my own and do not reflect those of my employer. This is shared for informational purposes only and should not be considered as official guidance or advice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9c914a341abd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>