gilice
765cd3823a
This is to some extent a simple transliteration of bench.sh, but also adds some additional features: - the ability to select which benchmarks to run, instead of having to run all at once (via --cases option) - help text (--help), and argument parsing - ability to profile instruction counts (via --mode icount). Instruction count is a stable metric, that correlates to the execution time, but is less affected by the environment (like other heavy programs running, or the system hardware). Other projects (most notably rustc) use instruction count as a primary target metric. Change-Id: I791cac499e64983efb8bcd4ea1faef13ded09ee6
Benchmarking scripts for Lix
These are very much WIP, and have a few clumsy assumptions that we would somewhat rather be fixed, but we have committed them to let others be able to do benchmarking in the mean time.
Benchmarking procedure
Build some Lixes you want to compare, by whichever means you wish.
Get a computer that is not busy and strongly preferably is bare-metal or at least not a cloud VM (e.g. go make coffee when running benchmarks).
From the root of a Lix checkout, run ./bench/bench.sh resultlink-one resultlink-two, where resultlink-one and resultlink-two are the result
links from the builds you want to test (they can be any directory with bin/nix
in it, however).
To get the summary again, run ./bench/summarize.jq bench/bench-*.json.
Example results
(vim tip: :r !bench/summarize.jq bench/bench-*.json to dump it directly into
your editor)
result-asserts/bin/nix --extra-experimental-features 'nix-command flakes' search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6
c6fe19ccb59cac54b5b3f25e160870 hello
mean: 15.993s ± 0.081s
user: 13.321s | system: 1.865s
median: 15.994s
range: 15.829s ... 16.096s
relative: 1
result/bin/nix --extra-experimental-features 'nix-command flakes' search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19cc
b59cac54b5b3f25e160870 hello
mean: 15.897s ± 0.075s
user: 13.248s | system: 1.843s
median: 15.88s
range: 15.807s ... 16.047s
relative: 0.994
---
result/bin/nix --extra-experimental-features 'nix-command flakes' eval -f bench/nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
mean: 0.4s ± 0.024s
user: 0.335s | system: 0.046s
median: 0.386s
range: 0.379s ... 0.43s
relative: 1
result-asserts/bin/nix --extra-experimental-features 'nix-command flakes' eval -f bench/nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
mean: 0.404s ± 0.024s
user: 0.338s | system: 0.046s
median: 0.386s
range: 0.384s ... 0.436s
relative: 1.008
---
result-asserts/bin/nix --extra-experimental-features 'nix-command flakes' eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
mean: 5.838s ± 0.023s
user: 5.083s | system: 0.464s
median: 5.845s
range: 5.799s ... 5.867s
relative: 1
result/bin/nix --extra-experimental-features 'nix-command flakes' eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
mean: 5.788s ± 0.044s
user: 5.056s | system: 0.439s
median: 5.79s
range: 5.715s ... 5.876s
relative: 0.991
---
GC_INITIAL_HEAP_SIZE=10g result-asserts/bin/nix eval --extra-experimental-features 'nix-command flakes' --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
mean: 4.147s ± 0.021s
user: 3.457s | system: 0.487s
median: 4.147s
range: 4.123s ... 4.195s
relative: 1
GC_INITIAL_HEAP_SIZE=10g result/bin/nix eval --extra-experimental-features 'nix-command flakes' --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
mean: 4.149s ± 0.027s
user: 3.483s | system: 0.456s
median: 4.142s
range: 4.126s ... 4.215s
relative: 1
---