Standalone MPI Worker

Deployable OpenMPI + gRPC worker for distributed epidemic simulation. Rank 0 serves gRPC; other ranks join the simulation and aggregate state with MPI_Allreduce. Designed to run on a separate machine from the middleware (see companion pandemic-sim orchestrator if you use the full stack).

Prerequisites

Docker (recommended)

• Docker Desktop (Windows/macOS) or Docker Engine + Compose (Linux)

Manual build (optional)

• CMake 3.15+
• OpenMPI
• gRPC and Protobuf
• C++17 compiler

Quick start (Docker)

git clone https://github.com/diptomistry/standalone-mpi-worker.git
cd standalone-mpi-worker

Optional .env (defaults listen on 0.0.0.0:50054 with 4 ranks):

WORKER_HOST=0.0.0.0
WORKER_PORT=50054
MPI_RANKS=4

docker-compose build
docker-compose up -d
docker-compose logs -f

Expected logs include rank initialization and Server listening on 0.0.0.0:50054.

Connecting to middleware

On the machine running middleware:

middleware:
  environment:
    - WORKER_MPI=192.168.1.60:50054  # MPI worker host:port

docker-compose restart middleware

How it works

• Container runs OpenMPI via mpirun -np MPI_RANKS ./mpi_worker <host:port>
• Ranks participate in the simulation; results are reduced with MPI_Allreduce
• Only rank 0 runs the gRPC server toward middleware

Layout

standalone-mpi-worker/
├── docker-compose.yml
├── Dockerfile
├── .env.example
├── CMakeLists.txt
├── proto/
│   ├── pandemic_sim.proto
│   └── CMakeLists.txt
└── workers/mpi/
    ├── CMakeLists.txt
    └── main.cpp

Security notes

• gRPC is without TLS (fine for local lab networks). Enable TLS for anything exposed beyond a trusted network.
• Restrict inbound TCP on WORKER_PORT to the middleware host.