Profiling

The Rust Performance Book lists a number of different profilers. I have used cargo-flamegraph and perf to profile navig18xx. This allowed me to identify that when finding the optimal pairing of trains to routes (Trains::select_routes) the majority of time was being spent determining whether pairs of paths had any conflicts or could be operated together. This analysis can be reproduced with the following steps:

1. Install the necessary packages; e.g., on Debian Linux:

   sudo apt install linux-perf
   cargo install flamegraph
   

2. Delete the cached routes for one company so that the 1867_bc example will have to identify them again:

   rm ./examples/output/1867_bc_D.json
   

3. Profile the 1867_bc example and generate a flamegraph:

   cargo flamegraph --example 1867_bc --output flamegraph-1867_bc.svg
   

Note: this can generate many gigabytes of performance data, particularly if you delete the cached routes for more than one company.

Collecting profiling information

Note that you may need to temporarily decrease the value of perf_event_paranoid in order to collect profiling information. You should then restore its original value.

echo 2 | sudo tee /proc/sys/kernel/perf_event_paranoid
cargo flamegraph --example 1867_bc --output flamegraph-1867_bc.svg
echo 3 | sudo tee /proc/sys/kernel/perf_event_paranoid

Improved output for release builds

If profiling a release build, you may want to ensure that full debugging information is collected so that the profiling output is easier to interpret. To do this, either set the environment variable CARGO_PROFILE_RELEASE_DEBUG=true or temporarily add the following lines to Cargo.toml:

[profile.release]
debug = true