How to make benchmarks run fine on windows?

[supercilex]

How to make benchmarks run fine on windows?

[SUPERCILEX]

As expected for windows right now since we're just forwarding to https://github.com/XAMPPRocky/remove_dir_all/blob/master/Cargo.toml. On that note, I noticed that the parallel feature is disabled by default. If you enable it here,

fuc/fuc_engine/Cargo.toml

Line 24 in cb34600

remove_dir_all = "0.8.2"

you should get about the same performance.

Can you try that and share benchmarks?

[SUPERCILEX]

You can use --show-output with hyperfine to see the actual failure. My guess: have you install ftzz?

[luceat-lux-vestra]

Thanks.

It seems hyperfine problem?
I can't pass command line options to commands.

I installed fizz and when I run this command on terminal

hyperfine --warmup 3 -N --show-output --prepare "ftzz g -n 100 -b 100 ""D:/tmp/ftzz""" "rmdir /S /Q D:/tmp/"
got this error.

Benchmark 1: rmdir /S /Q D:/tmp/
About 100 files will be generated in approximately 100 directories distributed across a tree of maximum depth 5 where each directory contains approximately 3 other directories. Each file will contain approximately 1 byte of random data.
Created 32 files (29 B) across 214 directories.
rmdir: failed to remove '/S': The system cannot find the file specified.
rmdir: failed to remove '/Q': The system cannot find the file specified.
rmdir: failed to remove 'D:/tmp/': The directory is not empty.
Error: Command terminated with non-zero exit code: 1. Use the '-i'/'--ignore-failure' option if you want to ignore this. Alternatively, use the '--show-output' option to debug what went wrong.

[luceat-lux-vestra]

Problem solved.
-N made options for command not usable.

Shell script converted to batch script on windows and it's working.
(It's from remote windows, line breaks not applied here.)

@echo off
for %%G in (0,1G) do (
 for %%F in (10,10_000,100_000,1M) do (
    hyperfine --warmup 3 ^
      --export-markdown "./benches/remove_%%F_files_%%G_bytes.md" ^
      --export-json "./benches/remove_%%F_files_%%G_bytes.json" ^
      --prepare "ftzz g -n %%F -b %%G D:/tmp/ftzz" ^
        "rmdir /S /Q ""D:/tmp""" ^
        "D:/Repositories/fuc/target/release/rm_stdlib D:/tmp/ftzz" ^
        "D:/Repositories/fuc//target/release/rm_rayon D:/tmp/ftzz" ^
        "D:/Repositories/fuc//target/release/rm_remove_dir_all D:/tmp/ftzz" ^
        "D:/Repositories/fuc//target/release/rmz D:/tmp/ftzz"
    )

  hyperfine --warmup 3 -N ^
    --export-markdown "benches/remove_100_000_files_%%G_bytes_0_depth.md" ^
    --export-json "benches/remove_100_000_files_%%G_bytes_0_depth.json" ^
    --prepare "ftzz g -n 100_000 -b %%G -d 0 D:/tmp/ftzz" ^
      "rmdir /S /Q ""D:/tmp""" ^
      "D:/Repositories/fuc/target/release/rm_stdlib D:/tmp/ftzz" ^
      "D:/Repositories/fuc/target/release/rm_rayon D:/tmp/ftzz" ^
      "D:/Repositories/fuc/target/release/rm_remove_dir_all D:/tmp/ftzz" ^
      "D:/Repositories/fuc/target/release/rmz D:/tmp/ftzz"

  hyperfine --warmup 3 -N ^
    --export-markdown "benches/remove_100_000_files_%%G_bytes_5_files_per_dir.md" ^
    --export-json "benches/remove_100_000_files_%%G_bytes_5_files_per_dir.json" ^
    --prepare "ftzz g -n 100_000 -b %%G -r 5 D:/tmp/ftzz" ^
      "rmdir /S /Q ""D:/tmp""" ^
      "D:/Repositories/fuc/target/release/rm_stdlib ""D:/tmp/ftzz""" ^
      "D:/Repositories/fuc/target/release/rm_rayon ""D:/tmp/ftzz""" ^
      "D:/Repositories/fuc/target/release/rm_remove_dir_all ""D:/tmp/ftzz""" ^
      "D:/Repositories/fuc/target/release/rmz ""D:/tmp/ftzz"""
)

[luceat-lux-vestra]

rm_rayon is fastest on windows 11(tested on HDD).

[luceat-lux-vestra]

@SUPERCILEX not relevant, but do you know some windows internals?

_Hm... I thought wrongly. _

I hope to get some advice on this.

Thanks.

[SUPERCILEX]

Sadly no, I don't know much of anything about Windows internals.

[luceat-lux-vestra]

benches.zip

I can't wait to finish remove benchmark running on HDD.
But I guess almost last phase.

need to adjust values or run benches on SSD...
HDD takes too much time.

[SUPERCILEX]

Thanks! So that's with the parallel feature enabled? Seems like a win for the large benchmarks.

[luceat-lux-vestra]

Yes, I enabled parallel feature for remove_dir_all.

[SUPERCILEX]

Sweet! I just pushed 844b20b

[luceat-lux-vestra]

benches (2).zip

Added copy benchmarks, just stdlib, rayon, cpz.
I'll try this with robocopy later.