Skip to main content

VRB Binderhub Server Deployment Guide with GPU Resources

Complete guide for deploying the Virtual Research Building (VRB) server — a cloud-based robotics platform built on the BinderHub project, with NVIDIA GPU support. BinderHub enables users to launch reproducible computing environments from Git repositories; this deployment extends it with GPU time-slicing and VirtualGL for GPU-accelerated rendering in user pods (Isaac Sim, OpenGL/Vulkan applications, etc.).

This guide covers two deployment targets, and each chapter walks through both in parallel:

TargetClusterGPU sharingPublic access
Google CloudGKE managed cluster with GPU node poolsGKE-native time-slicing (set at node-pool creation via gcloud flags)GKE Load Balancer + nginx ingress + cert-manager (Let's Encrypt)
Self-hostedMicroK8s on your own Ubuntu machineNVIDIA gpu-operator + time-slicing-config-all.yamlCloudflare Tunnel

Pick whichever fits your situation — the chapters mark which steps apply to which target.

Architecture Overview​

flowchart TB
    User([User Browser])

    subgraph Public["Public Network"]
        DNS[DNS / HTTPS<br/>binder.example.org<br/>jupyter.example.org]
    end

    subgraph Ingress["Ingress Layer"]
        LB1[BinderHub<br/>LoadBalancer]
        LB2[JupyterHub proxy-public<br/>LoadBalancer]
    end

    subgraph K8s["Kubernetes Cluster (GKE or MicroK8s)"]
        subgraph BinderNS["namespace: binder"]
            BH[BinderHub Pod<br/>build & launch API]
            Hub[JupyterHub Pod<br/>user auth & spawning]
            Proxy[Proxy Pod<br/>traffic routing]
            Sched[User Scheduler]
            UserPods[User Pods<br/>JupyterLab + VirtualGL<br/>GPU via time-slicing]
        end

        subgraph GPU["GPU Layer"]
            GPUPlugin[NVIDIA Device Plugin<br/>time-slicing: 2 vGPUs / GPU]
            PhysGPU[(Physical GPU)]
        end
    end

    Registry[(Container Registry<br/>Docker Hub / GCR)]
    GitRepo[(Git Repository<br/>GitHub)]

    User --> DNS
    DNS --> LB1
    DNS --> LB2
    LB1 --> BH
    LB2 --> Proxy
    Proxy --> Hub
    Hub --> Sched
    Sched --> UserPods
    BH -->|pull source| GitRepo
    BH -->|build & push| Registry
    UserPods -->|pull image| Registry
    UserPods --> GPUPlugin
    GPUPlugin --> PhysGPU

Table of Contents​

ChapterDescription
1. Kubernetes Cluster SetupSet up the Kubernetes cluster (self-hosted MicroK8s or Google Cloud GKE)
2. Deploy BinderHubPrepare host filesystem, deploy BinderHub via Helm, expose to public network
3. GPU Time-SlicingConfigure NVIDIA GPU time-slicing for multi-pod GPU sharing
4. Troubleshootingtroubleshooting, Day-to-day operations
5. Shutdown and UninstallTear down services and clean up resources

References:​