The containers-benchmark package

[Tags: bsd3, program]

This package evaluates performance of a containers package from multiple perspectives.

It measures time and memory allocation of methods specialised to several types. It can also measure the size and memory allocation of GHC compiled with the containers package.

The containers package used can be both system-installed or any darcs repository.


[Skip to ReadMe]

Properties

Versions1.0, 1.0.0.1, 1.0.0.2, 1.0.0.3, 1.1.0.0
Change logNone available
Dependenciesbase (>=3 && <5), bytestring (>=0.9 && <1.0), containers (>=0.3 && <0.5), criterion (==0.5.*), random (==1.0.*) [details]
LicenseBSD3
AuthorMilan Straka
Maintainerfox@ucw.cz
StabilityExperimental
CategoryBenchmarking
Home pagehttp://fox.auryn.cz/darcs/containers-benchmark/
Executablesbenchmark
UploadedSun Oct 31 07:00:32 UTC 2010 by MilanStraka
DistributionsNixOS:1.1.0.0
Downloads838 total (40 in last 30 days)
Votes
0 []
StatusDocs not available [build log]
All reported builds failed as of 2015-05-26 [all 1 reports]

Downloads

Maintainers' corner

For package maintainers and hackage trustees

Readme for containers-benchmark-1.0

This package consists of a thorough benchmark for a containers package.

Quick start
===========
To benchmark the installed version of containers, just do

cd scripts
sh benchmark.sh installed-containers none

Running
=======

The benchmark is executed using scripts in scripts directory. The main
executable is benchmark.sh, which is given several arguments:
benchmark.sh
  name_of_the_benchmark      -- date is added to it
  containers_repo            -- repo with containers, write 'none' for
                             --   system containers library
  cabal_configure_opts       -- optional flags passed to cabal configure

The benchmark.sh just calls other scripts:
  compile.sh containers_repo cabal_configure_opts   -- builds the benchmark
  run.sh name_of_the_benchmark                      -- run built benchmark
  collect.pl name_of_benchmark      -- collect all results in one .csv file

Other scripts:
  see.sh < name_of_csv              -- formats the csv file for viewing
  cmp.pl base_csv diff_csv          -- generate csv with comparison of two tests
  cmp.sh base_csv diff_csv          -- formats the result of cmp.pl for viewing

  ghc.sh containers_repo ghc_repo result_csv
    -- Compiles the GHC repo with given containers and store results
    -- the size of GHC and the memory allocation results of performance tests
    -- are stored. These tests were once a regression, so we monitor them.
    -- The GHC repo must be already booted and configured.


Which implementation to benchmark
=================================

The benchmark can use system-installed containers package or any given repo.
Because the criterion package used for benchmarking is using containers
internally, we cannot safely expose the benchmarked implementation as
Data.Map and others. Therefore we benchmark Container.{Map,Set,IntMap,IntSet}.
The scripts/system_containers.sh and scripts/custom_containers.sh scripts
chooses which implementation of containers is benchmarked.


Benchmark design
================

Various input data are provided by InputData/*

Various benchmark methods are provided by Benchmar/*

As the different combinations of input data and containers result in different
type signatures, we use CPP extension to put the input data and benchmark
methods into one source file and let GHC to type-check for each combination of
input data and container. This happens in Variants/*