Elastic AI computation

DSS can scale most of its processing by pushing down computation to Elastic computation clusters powered by Kubernetes

• Concepts
  • Interaction between DSS and containers
  • Capabilities and benefits
    • Limitations
  • Containerized execution configurations
  • Spark configurations
  • Execution configurations vs clusters
  • Base images
  • Support of code environments
• Initial setup
  • Prerequisites
    • Docker and kubectl setup
    • Other prerequisites include
  • (Optional) Setup Spark
  • Build the base image
  • (Optional) Build the Spark base image
  • Setting up containerized execution configs
  • Setting up Spark configurations
  • Push Base images
  • Use Kubernetes
• Managed Kubernetes clusters
  • Creating a cluster
  • Using the cluster
  • Advanced usage for multiple managed clusters
    • Use a specific or dynamic cluster for scenarios
      • Use a specific static cluster
      • Use a dynamic cluster
    • Automate start and stop of clusters
    • Permissions
• Using Amazon Elastic Kubernetes Service (EKS)
  • Using managed EKS clusters
    • Initial Setup
      • Install the EKS plugin
      • Prepare your local commands
      • Create base images
      • Create a new containerized execution configuration
    • Cluster configuration
      • Connection
      • Network settings
      • Cluster nodes
    • Using GPUs
      • Building an image with CUDA support
      • Enable GPU support on the cluster
      • Add a custom reservation
      • Deploy
  • Using unmanaged EKS clusters
    • Setup
      • Create your EKS cluster
      • Prepare your local aws, docker, and kubectl commands
      • Create base images
      • Create a new execution configuration
    • Using GPUs
      • Building an image with CUDA support
      • Enable GPU support on the cluster
      • Add a custom reservation
      • Deploy
• Using Microsoft Azure Kubernetes Service (AKS)
  • Using managed AKS clusters
    • Initial setup
      • Create your ACR registry
      • Install the AKS plugin
      • Prepare your local az, docker, and kubectl commands
      • Create base images
      • Create a new containerized execution configuration
    • Cluster configuration
    • Using GPUs
      • Building an image with CUDA support
      • Enable GPU support on the cluster
      • Add a custom reservation
      • Deploy
    • Other
  • Using unmanaged AKS clusters
    • Setup
      • Create your ACR registry
      • Create your AKS cluster
      • Prepare your local az, docker, and kubectl commands
      • Create base images
      • Create a new containerized execution configuration
    • Using GPUs
      • Building an image with CUDA support
      • Create configuration and add a custom reservation
      • Create a cluster with GPUs
      • Deploy
• Using Google Kubernetes Engine (GKE)
  • Using managed GKE clusters
    • Initial setup
      • Install the GKE plugin
      • Prepare your local commands
      • Create base images
      • Create a new execution configuration
    • Cluster configuration
      • Connection
      • Network settings
      • Cluster nodes
    • Using GPUs
      • Building an image with CUDA support
      • Enable GPU support on the cluster
      • Add a custom reservation
      • Deploy
  • Using unmanaged GKE clusters
    • Setup
      • Create your GKE cluster
      • Prepare your local gcloud, docker, and kubectl commands
      • Create base images
      • Create the execution configuration
    • Using GPUs
      • Building an image with CUDA support
      • Enable GPU support on the cluster
      • Add a custom reservation
      • Deploy
• Using code envs with containerized execution
  • Code environment resources directory
  • Updating code envs
• Containerized DSS engine
  • Setup
    • Build the base image
    • Instance-specific image
    • Plugins
    • Containerized execution configs
  • Running a visual recipe in a container
  • Tuning
• Dynamic namespace management
  • Namespace policies
• Customization of base images
  • Building an image with CUDA support
    • Multiple base images
    • Setting a proxy
    • Adding system packages
    • Add a Dockerfile fragment
    • Completely custom Dockerfile
• Unmanaged Kubernetes clusters
  • Using a single unmanaged cluster
  • Using multiple unmanaged clusters
• Using Openshift
• Using NVIDIA DGX Systems
• Troubleshooting
  • Jobs fail to run
    • requests.exceptions.ConnectionError
    • Kubernetes job failed, exitCode=1, reason=Error
    • Spark on Kubernetes
• Using Docker instead of Kubernetes
  • Why Kubernetes rather than Docker
    • Kubernetes
    • Docker
  • Prerequisites (Docker)
  • Other prerequisites
  • Build the base image
  • Running in Docker
  • Remote daemons
    • Rationale
    • Setup
      • With a registry
      • Without a registry
    • Usage
  • Containerized execution configurations
    • Multiple execution configurations