rubyx/test/bench/results.md

# Benchmarks

hello - output hello world to measure kernel calls
add   - run integer adds by linear fibonacci of 20
call  - exercise calling by recursive fibonacci of 10
noop  - a baseline that does nothing

All programs (apart from noop) run 100k times 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.
# 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
to 100k.

The machine was a virtual arm (qemu) run on a acer swift 5 (i5 8265 3.9GHz), performance roughly equivalent to a raspberry pi.
But results (in ms) should be seen as relative, not absolute.


language  |  noop   |  hello  |  add    | call        
c         |  45     |  100    |  72     | 591
go        |  53     |  4060   |  64     | 624
rubyx     |  42     |  1245   | ?????   | ????

ruby      |  1830   | 2750    | 3000    | 1900_000

Comparison with ruby, not really for speed, just to see how much leeway there is in the future.
thats most of the benchmarking strange about the puts in soml, hard to see where a factor of 10 would come from 2015-11-20 22:51:58 +01:00			`# Benchmarks`

updating benchmarks on the new machine, consistent not rubyx yet 2019-07-23 19:14:28 +02:00			`hello - output hello world to measure kernel calls`
			`add - run integer adds by linear fibonacci of 20`
			`call - exercise calling by recursive fibonacci of 10`
			`noop - a baseline that does nothing`
thats most of the benchmarking strange about the puts in soml, hard to see where a factor of 10 would come from 2015-11-20 22:51:58 +01:00
fixing test to pass and extract own kind of same tests to bench bench tests will need some loop, but not as large as c 2019-07-25 20:25:15 +02:00			`All programs (apart from noop) run 100k times to minimize startup impact.`
thats most of the benchmarking strange about the puts in soml, hard to see where a factor of 10 would come from 2015-11-20 22:51:58 +01:00
updating benchmarks on the new machine, consistent not rubyx yet 2019-07-23 19:14:28 +02:00			`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.`
thats most of the benchmarking strange about the puts in soml, hard to see where a factor of 10 would come from 2015-11-20 22:51:58 +01:00			`# Results`

			`Results were measured by a ruby script. Mean and variance was measured until variance was low,`
fixing test to pass and extract own kind of same tests to bench bench tests will need some loop, but not as large as c 2019-07-25 20:25:15 +02:00			`always under one percent. Noop showed that program startup is a factor, so all programs loop`
			`to 100k.`
thats most of the benchmarking strange about the puts in soml, hard to see where a factor of 10 would come from 2015-11-20 22:51:58 +01:00
updating benchmarks on the new machine, consistent not rubyx yet 2019-07-23 19:14:28 +02:00			`The machine was a virtual arm (qemu) run on a acer swift 5 (i5 8265 3.9GHz), performance roughly equivalent to a raspberry pi.`
			`But results (in ms) should be seen as relative, not absolute.`
did the ratios i think percentages with +/- is best otherwise it becomes an issue of which way you look at it. 1,33 or inverse 0.75 looks like a third one way and a quarter the other 2015-11-21 16:04:20 +01:00

updating benchmarks on the new machine, consistent not rubyx yet 2019-07-23 19:14:28 +02:00			`language \| noop \| hello \| add \| call`
			`c \| 45 \| 100 \| 72 \| 591`
			`go \| 53 \| 4060 \| 64 \| 624`
fixing test to pass and extract own kind of same tests to bench bench tests will need some loop, but not as large as c 2019-07-25 20:25:15 +02:00			`rubyx \| 42 \| 1245 \| ????? \| ????`
add ruby bench programs and numbers 2015-11-24 14:33:16 +01:00
updating benchmarks on the new machine, consistent not rubyx yet 2019-07-23 19:14:28 +02:00			`ruby \| 1830 \| 2750 \| 3000 \| 1900_000`
add ruby bench programs and numbers 2015-11-24 14:33:16 +01:00
fixing test to pass and extract own kind of same tests to bench bench tests will need some loop, but not as large as c 2019-07-25 20:25:15 +02:00			`Comparison with ruby, not really for speed, just to see how much leeway there is in the future.`