Performance benchmarking

[mgree]

Timing: use https://crates.io/crates/tracing-timing?

Microbenchmarks

• read
• write

Macrobenchmarks

• tasks?

Compare against

• FS (various hdd configs) as baseline
• ramdisk as memory FS baseline
• jq
• js
• python
• native rust

[mgree]

To FUSE or Not to FUSE: Performance of User-Space File Systems for background data on FUSE performance.

High Velocity Kernel File Systems with Bento for a Rust-y take on FUSE (direct VFS bindings in C call a safe FS API in Rust)

[mgree]

I can imagine several abstract benchmarks:

• [ ] Time to read a single file in the root
• [ ] Time to read a single file nested (deeply/widely)
• [ ] Time to read k randomly (predetermined) files out of N >> k files

Each of the above should also have writing/creation tests.

• [ ] Time to sum every numeric entry in a large file (deep or wide)

And several benchmarks that could run on real data:

• [ ] Time to sum every, e.g. cost entry in a long list of records
• [ ] Time to find commits matching a given pattern from a GH dump
• [ ] Time to generate, e.g., a GraphQL call or a build/CI script
• [ ] Time to modify, e.g., a GraphQL call