How um-bridge/umbridge works
Project type and purpose
• UM-Bridge is a protocol-driven, language-agnostic “model-server” framework.
• It exposes numerical models (C++, Python, Julia, etc.) via HTTP so that any client (R, Python, C++, MATLAB, Julia, Muq) can call them in a uniform way.
• It also provides deployment scripts (HPC, Kubernetes), clients in many languages, CI workflows, benchmarks, and protocol‐conformance tests.
File structure
• Top-level “umbridge/” directory: the core Python package implementing the client-side protocol.
• “clients/” subfolders: example clients in R, C++, Python, MATLAB, Julia, Muq.
• “models/”: template minimal servers in C++, Python, Julia plus Dockerfiles.
• “hpc/”: LoadBalancer code, SLURM scripts, helper scripts for high-throughput cluster deployment.
• “kubernetes/”: YAMLs and launch scripts for containerized, multi-node, MPI model deployments.
• “benchmarks/”: testbenchmark server and Dockerfile for UQ benchmarking.
• “testing/”: protocol-conformity tests for versions 0.9 and 1.0, plus client‐side unit tests.
• CI workflows in “.github/workflows” orchestrate build, test, documentation, and multi-language client package releases.
• “R/” subtree: R wrapper package.
• “lib/”: low-level HTTP utility headers (httplib, JSON, umbridge.h).
• “setup.py”: Python packaging entry point.
Key architecture insights from README
• UM-Bridge’s core goal is “coupling advanced models … to statistical or optimization methods” via a unified HTTP interface.
• It is designed as a modular, client–server system with clearly defined protocol versions.
Final advice
• Break the overall system into these subsystems and show how they interconnect.
• Be exhaustive: include each language client, each deployment mode, each test/benchmark component.