forked from mirrors/jj
a03e9c41a3
Some benchmark numbers in the following order:
- original
- fresh repo, BatchSize::SmallInput
- fresh but index-preloaded repo, BatchSize::SmallInput
- fresh but index-preloaded repo, BatchSize::LargeInput
% cargo run --release --features bench -- bench revset ':main'
revsets/:main time: [271.49 µs 271.74 µs 272.07 µs]
revsets/:main time: [754.17 µs 758.22 µs 764.02 µs]
revsets/:main time: [367.11 µs 372.65 µs 381.99 µs]
revsets/:main time: [341.76 µs 342.98 µs 344.35 µs]
% cargo run --release --features bench -- bench revset 'author(martinvonz)'
revsets/author(martinvonz)
time: [767.43 µs 770.52 µs 775.59 µs]
revsets/author(martinvonz)
time: [31.960 ms 31.984 ms 32.011 ms]
revsets/author(martinvonz)
time: [31.478 ms 31.538 ms 31.615 ms]
revsets/author(martinvonz)
time: [31.503 ms 31.526 ms 31.550 ms]
I think the fresh but index-preloaded repo is close to the practical
evaluation environment. With BatchSize::SmallInput, it appears to consume
~600MB (RES) memory (compared to ~50MB in LargeInput.) I don't think that's
huge, but it might affect cache usage, so I chose LargeInput.
https://docs.rs/criterion/latest/criterion/enum.BatchSize.html
|
||
|---|---|---|
| .. | ||
| commands | ||
| config | ||
| cleanup_guard.rs | ||
| cli_util.rs | ||
| commit_templater.rs | ||
| config-schema.json | ||
| config.rs | ||
| diff_util.rs | ||
| formatter.rs | ||
| graphlog.rs | ||
| lib.rs | ||
| main.rs | ||
| merge_tools.rs | ||
| operation_templater.rs | ||
| progress.rs | ||
| template.pest | ||
| template_builder.rs | ||
| template_parser.rs | ||
| templater.rs | ||
| text_util.rs | ||
| time_util.rs | ||
| ui.rs | ||