Criterion allows us to crank up the sampling size and confidence, which is important for these kinds of microbenchmarks that can have high variance. It also has other bells and whistles like tracking relative performance between runs, which can be very helpful for development, and no longer requires a nightly compiler to run the benchmarks.
Per discussion in Discord, the presence of pico in these benchmarks has outlived its usefulness; we're now primarily concerned with the relative performance of httparse between changes.
Criterion allows us to crank up the sampling size and confidence, which is important for these kinds of microbenchmarks that can have high variance. It also has other bells and whistles like tracking relative performance between runs, which can be very helpful for development, and no longer requires a nightly compiler to run the benchmarks.
Per discussion in Discord, the presence of pico in these benchmarks has outlived its usefulness; we're now primarily concerned with the relative performance of httparse between changes.