rubyx/test/bench/results.md
Torsten Rüger d3a0aa3d8e redid benchmarks with new options
especially calling did improve
also binaries don't get crazy big even for bigger heaps, because they are just ints
2019-08-24 16:03:35 +03:00

31 lines
1.3 KiB
Markdown

# Benchmarks
hello - output hello world to measure kernel calls
add - run integer adds by linear fibonacci of 40
call - exercise calling by recursive fibonacci of 20
noop - a baseline that does nothing
loop - just counts down, from 1M
Loop, Hello, add and call run 1M , 50k, 10k and 100 respectively,
to minimize startup impact.
C was linked statically as dynamic linked influences times.
Output was sent to /dev/null, so as to measure the calling and not the terminal.
Also output was unbuffered, because that is what rubyx implements.
# Results
Results were measured by a ruby script. Mean and variance was measured until variance was low,
always under one percent. Noop showed that program startup is a factor, so all programs loop somewhere from 1M to 100, depending on how intensive.
The machine was a virtual arm (qemu) run on a acer swift 5 (i5 8265 3.9GHz), performance roughly equivalent to a raspberry pi.
Results (in ms) should be seen as relative, not absolute.
language | noop | hello | add | call | loop
c | 55 | 330 | 88 | 135 | 6
go | 52 | 400 | 9 | 77 | 2
rubyx | 42 | 160 | 1575 | 780 | 430
ruby | 1570 | 650 | 1090 | 1500 | 180
mruby | 86 | 1100 | 1370 | 2700 | 300