The FHIR landscape is expanding and maturing rapidly. Users now have many great FHIR server implementation options to choose between. This project aims to help with those choices by reporting on the performance of FHIR server implementations, via a set of repeatable and isolated benchmarks that have captured:

• The throughput of various operations, i.e. responses per second.
• The latency of various operations, i.e. how many milliseconds per response, at various percentiles, e.g. "100% of responses take less than 42 milliseconds", "99.9% of responses take less than 30 milliseconds", "50% of responses take less than 10 milliseconds (i.e. the median latency)", etc.
• How those operations performed under different levels of load/concurrency.
• How those operations performed with sample data sets of different sizes.
• The success-to-failure ratio of those operations.

Each round of benchmarking will be run periodically and published publicly to fhir-benchmarks.com, allowing users to see the relative performance of various FHIR servers and FHIR server implementers to gauge (and improve!) their server's performance.

Some of the most immediate next steps include:

• Adding support for benchmarking additional operations, e.g. Patient writes, reads, and searches.
• Adding support for additional FHIR servers, e.g. IBM FHIR Server.
• Adding infrastructure and automation to run the benchmarks periodically and automatically.

The benchmark orchestrator and operations are all written in Rust, which run against locally hosted FHIR servers run as Docker containers, using sample data generated via Synthea.

First, ensure you have installed the following prerequisites:

• Rust >= v1.51.0, as installed via rustup.
• Docker >= v20.10.6
• Docker Compose >= v1.27.4
• On Ubuntu 20.04, these additional steps are also required:

  • Install some additional dependencies:

    $ sudo apt install libssl-dev pkg-config
    

  • Give the current user permissions to run docker without sudo:

    $ sudo usermod -aG docker $USER
    

Then, run these commands to clone, build, and run the benchmark suite's tests:

$ git clone https://github.com/karlmdavis/fhir-benchmarks.git
$ cd fhir-benchmarks
$ # Pull in the submodules:
$ git submodule update --init --recursive
$ # Build in debug mode and run tests:
$ cargo test
$ # Build in release mode and run benchmarks:
$ cargo run --release \
  | tee ./results/results-release-$(date -u +"%Y-%m-%dT%H:%M:%SZ").json

Both the tests and the benchmarks themselves will automatically build the Docker containers for the FHIR servers and Synthea, as needed.

Or to run the benchmarks with a customized config (see the defaults in ./fhir-bench-orchestrator/src/config.rs):

$ FHIR_BENCH_ITERATIONS=1000 \
  FHIR_BENCH_CONCURRENCY_LEVELS=1,2,8 \
  FHIR_BENCH_POPULATION_SIZE=1000 \
    cargo run --release \
    | tee ./results/benchmark-$(date -u +"%Y-%m-%dT%H:%M:%SZ").json

VS Code is hard to beat as an IDE for Rust projects like this one. If you'd like to use it, I'd suggest installing the following extensions:

• rust-analyzer: This extension adds Rust support to VS Code. It's honestly pretty amazing, particularly in how it will mark up your code with helpful type annotations.
• crates: Makes it easy to see if Cargo.toml files, which are used to specify the dependencies for a Rust project, are specifying the latest versions of each dependency. Simple, but major quality of life improvement.

You'll want to ensure that your ~/Library/Application Support/Code/User/settings.json file, which specifies the settings for VS Code, contains the following:

{
    "rust-analyzer.runnableEnv": {
        "RUST_BACKTRACE": "full",
        "RUST_LOG": "info",
        "RUST_LOG_SPAN_EVENTS": "new,close"
    }
}

If you'd like to add support for a new FHIR server, additional operations, or otherwise help to build out these benchmarks, you'd be very welcome! Please see CONTRIBUTING.md for further details.

This project is in the worldwide public domain. As stated in CONTRIBUTING:

This project is in the public domain within the United States, and copyright and related rights in the work worldwide are waived through the CC0 1.0 Universal public domain dedication.

All contributions to this project will be released under the CC0 dedication. By submitting a pull request, you are agreeing to comply with this waiver of copyright interest.