non-trivial-benchmarks's Introduction

non-trivial-benchmarks

Some Common lisp benchmarks to compare libraries and techniques.

Setup

If you would like to run or adapt the test suite it's best you use qlot. Qlot provides a way for developers to ensure they have the library versions as the last time the tests where ran.

(Option 1) Roswell

Installing qlot with roswell is as easy as

ros install qlot

Then ensure qlot is in your path

Then install the dependencies using

qlot install

(Option 2) Quicklisp

Simply start up a repl at the root directory then in the repl run

(ql:quickload :qlot)
(qlot:install :non-trivial-benchmarks)

Running the benchmarks

If the dependencies were installed correctly you can run the benchmarks by running make in the root directory.

How does non-trivial-benchmarks differ from cl-bench?

cl-bench contains benchmarks aimed at comparing the performances of different common lisp implementations, whereas
non-trivial-benchmarks aims to compare different techniques, methodologies, algorithms, etc within the same lisp implementation.

non-trivial-benchmarks's People

Contributors

Stargazers

Watchers

non-trivial-benchmarks's Issues

use qlot to pin benchmarked projects

Readme: explain the difference(s) between this and cl-bench

add the "string building" benchmarks from @fstamour's "catchall" repo

These benchmarks: https://gitlab.com/fstamour/catchall/-/blob/c78e0fe2888024ab8a905fc187a5760322179195/string-builder.lisp

brainstorm on things to benchmark

I should create issues for all these... Maybe I could create a "brainstorm" issues to dumps all ideas in it, and later create specific issues as needed

provide a way to select which benchmarks to run

this is a must

Add benchmark for using sbcl's native message box and queue implementations

You can find them documented in http://www.sbcl.org/manual/#sb_002dconcurrency I think they could be put in the same section as the channel implementations. Specifically message boxes which to many people are just like channels.

plot some graphs

perhaps using gnuplot?

We first need to extract the data from the benchmarks.
Then we should save that data to disk. This will save time during development, and would - eventually - let other people analyze the data without having to run the benchmarks.
Load the data
Plot it!