search

krlmlr / dplyr.benchmark

dplyr benchmarks
https://krlmlr.github.io/dplyr.benchmark
2 stars 2 forks source link

Improve calibration #3

Open krlmlr opened 7 years ago

krlmlr commented 7 years ago

The current calibration is insufficient, for now all runs should be made on the same machine. This impedes automated benchmarking, e.g., on Travis.

A benchmark for the whole history can be rerun in a matter of hours (on 16 cores), so even if the benchmark machine changes this is not much of a problem.

krlmlr commented 7 years ago

I had locally -O0, pretty sure this has skewed the results a lot. Need to recheck without that.