Documentation

Gathers documentation on setting up UM-Bridge models, clients and running using HPC systems or cloud platforms.

• Models
  • Test model
  • Implementation
    • Python server
    • C++ server
    • Julia server
  • MUQ server
• Clients
  • Python client
  • C++ client
  • R client
  • Matlab client
  • Julia client
  • MUQ client
  • PyMC client
  • QMCPy client
  • tinyDA client
  • emcee client
  • TT-Toolbox client
  • Sparse Grids Matlab Kit client
• Google Kubernetes Engine
  • Step 1: Install cloud shell
  • Step 2: Initialise and run cloud shell
  • Step 3: Clone UM-Bridge
  • Step 4: Initialise a cluster
  • Step 5: Set up UM-Bridge kubernetes cluster
• Kubernetes
  • Step 1: Clone UM-Bridge
  • Step 2: Set up load balancer
  • Step 3: Run model instances
  • Step 4: Calling the model
  • Multinode MPI on kubernetes
    • Step 1: mpi-operator base image
    • Step 2: Deploy mpi-operator
    • Step 3: Setting up NFS
    • Step 4: Running a job on the new cluster
• Interface
  • Protocol
    • Verifying correctness
    • Required endpoints
    • Optional endpoints
    • POST /InputSizes
    • POST /OutputSizes
    • GET /Info
    • POST /ModelInfo
    • POST /Evaluate
    • POST /Gradient
    • POST /ApplyJacobian
    • POST /ApplyHessian
    • Errors
• HPC
  • Installation
  • Usage
    • Choosing a suitable scheduler
    • (Option 1) SLURM scheduler
    • (Option 2) HyperQueue scheduler
    • Running the load balancer
  • Resource management with HyperQueue
    • Specifying HyperQueue worker resources
    • Requesting hardware resources with a job manager (SLURM/PBS)
  • (Optional) Launching MPI tasks with HyperQueue under SLURM
    • Support for multi-node tasks in HyperQueue
    • Launching MPI tasks in a HyperQueue job under SLURM
  • (Optional) Varying resource requirements per model (e.g. for multilevel / multifidelity)
  • (Optional) Running clients on your own machine while offloading runs to HPC
• Benchmarks
  • Test benchmark
  • Contributing
    • Dockerfile
    • README
• Launching containerised MPI applications in HPC
  • Docker to Singularity
    • If you have –fakeroot
    • If you don’t have –fakeroot
    • Running Singularity containers
  • MPI with Singularity + UM-Bridge
    • Using the Hybrid model method
    • Using the container “as is”
• Mathematical abstraction in UM-Bridge
  • Model Evaluation
  • Gradient of the objective function
  • Applying Jacobian to a vector
  • Applying Hessian to a vector