Curl Performance

Love the idea. Some suggestions from a person who's a bit too much into graphs:

The axes labelling could use some work - what's 3.5G on y? What's R on x? (Yes, rounds, but how does it map to time/commits?) Even with the description available above the graph, it would be great if a copy of that image provided all basic information.

GBps on top of the axis would be way better than G by each number.

A high ceiling for "what can iperf achieve" would be good to know. (Is curl already at the top, or is there a 50% improvement for grabs)

Instead of hiding away the distribution, it could be vertical, on the left of the y axis.

I really like that they included some round/machine-independent (mostly) options like allocations and memory use. Number of instructions retired and cache misses could be also interesting.

And it was totally a good approach to ignore grafana. It's awesome software, but entirely unnecessary if you have static data. Even at a million lines, you can easily scan them and produce a graph in under a second. And getting to choose your own specific display style makes a difference. If you have some more fancy querying to add, you could even try CDB, sqlite, duckdb or some simple storage.

The JS niceity I'd like would be a way to highlight a specific series on the graph where they interleave a lot.

The Python package Pandas uses the package Airspeed Velocity to measure performance over time: https://pandas.pydata.org/community/benchmarks.html

Actually surprised the data is not garbage from random stuff running on a machine running a GUI.

But last time I did this kind of extensive testing was over 10, maybe even 15 years ago and it really made everything easier to just put a dedicated box with near zero services there - like a NUC.