search

mbrubeck / tree_magic

Determines the MIME type of a file by traversing a filetype tree.
MIT License
34 stars 9 forks source link

Improve performance by removing `fnv` dependency #21

Open aehancoc opened 6 months ago

aehancoc commented 6 months ago

Hey, long time no see. Thanks for maintaining this in my absence. (In case you're wondering, my old account was tied to my university email, which I lost access to.)

Anywho, I chose fnv back in https://github.com/mbrubeck/tree_magic/commit/b93b80f8d39796a389d90d5686dfa80f100275cf because std::collections::HashMap had somewhat worse performance. It seems things have changed, and now std's hashmap/hashset is better now. fnv is also really terrible with larger keys, which we're dealing with here due to using strings to store MIME types. I also took the liberty of adding the new test files to the benchmarks and removing the duplicate files there.

Some numbers from my Intel Core i5-1135G7:

fnv

     Running benches\from_u8.rs (target\release\deps\from_u8-ff4f06eaaf1dbfd9.exe)

running 18 tests
test application_x_7z_compressed ... bench:       3,103 ns/iter (+/- 409)
test application_x_tar           ... bench:      23,813 ns/iter (+/- 6,375)
test application_zip             ... bench:       6,430 ns/iter (+/- 1,344)
test audio_flac                  ... bench:      14,739 ns/iter (+/- 445)
test audio_mpeg                  ... bench:      21,131 ns/iter (+/- 3,454)
test audio_ogg                   ... bench:       6,906 ns/iter (+/- 337)
test audio_opus                  ... bench:       7,053 ns/iter (+/- 965)
test audio_wav                   ... bench:       7,615 ns/iter (+/- 818)
test image_bmp                   ... bench:      12,934 ns/iter (+/- 3,783)
test image_gif                   ... bench:       2,315 ns/iter (+/- 662)
test image_png                   ... bench:       2,448 ns/iter (+/- 1,183)
test image_tiff                  ... bench:      20,376 ns/iter (+/- 1,384)
test image_x_pcx                 ... bench:      13,892 ns/iter (+/- 3,323)
test image_x_portable_bitmap     ... bench:      14,249 ns/iter (+/- 3,162)
test image_x_tga                 ... bench:      10,709 ns/iter (+/- 1,521)
test image_xbm                   ... bench:      32,445 ns/iter (+/- 6,046)
test text_html                   ... bench:      29,509 ns/iter (+/- 4,487)
test text_plain                  ... bench:      28,937 ns/iter (+/- 6,604)

test result: ok. 0 passed; 0 failed; 0 ignored; 18 measured

     Running benches\match_u8.rs (target\release\deps\match_u8-bea7a4ade2ef60c5.exe)

running 18 tests
test application_x_7z_compressed ... bench:          84 ns/iter (+/- 17)
test application_x_tar           ... bench:          64 ns/iter (+/- 24)
test application_zip             ... bench:          60 ns/iter (+/- 13)
test audio_flac                  ... bench:          44 ns/iter (+/- 9)
test audio_mpeg                  ... bench:          52 ns/iter (+/- 4)
test audio_ogg                   ... bench:          41 ns/iter (+/- 6)
test audio_opus                  ... bench:          18 ns/iter (+/- 9)
test audio_wav                   ... bench:          58 ns/iter (+/- 9)
test image_bmp                   ... bench:          54 ns/iter (+/- 15)
test image_gif                   ... bench:          39 ns/iter (+/- 27)
test image_png                   ... bench:          46 ns/iter (+/- 5)
test image_tiff                  ... bench:          56 ns/iter (+/- 5)
test image_x_pcx                 ... bench:          91 ns/iter (+/- 35)
test image_x_portable_bitmap     ... bench:         134 ns/iter (+/- 117)
test image_x_tga                 ... bench:          61 ns/iter (+/- 100)
test image_xbm                   ... bench:          23 ns/iter (+/- 15)
test text_html                   ... bench:          90 ns/iter (+/- 32)
test text_plain                  ... bench:          36 ns/iter (+/- 21)

test result: ok. 0 passed; 0 failed; 0 ignored; 18 measured

std::collections

     Running benches\from_u8.rs (target\release\deps\from_u8-738934f266f65ebd.exe)

running 18 tests
test application_x_7z_compressed ... bench:      13,210 ns/iter (+/- 1,699)
test application_x_tar           ... bench:      20,664 ns/iter (+/- 4,171)
test application_zip             ... bench:       5,496 ns/iter (+/- 1,352)
test audio_flac                  ... bench:       9,246 ns/iter (+/- 5,188)
test audio_mpeg                  ... bench:       9,043 ns/iter (+/- 3,767)
test audio_ogg                   ... bench:       4,578 ns/iter (+/- 5,994)
test audio_opus                  ... bench:       4,980 ns/iter (+/- 2,177)
test audio_wav                   ... bench:       7,923 ns/iter (+/- 4,984)
test image_bmp                   ... bench:       8,892 ns/iter (+/- 4,375)
test image_gif                   ... bench:       2,219 ns/iter (+/- 742)
test image_png                   ... bench:       2,192 ns/iter (+/- 810)
test image_tiff                  ... bench:      10,511 ns/iter (+/- 6,045)
test image_x_pcx                 ... bench:       4,581 ns/iter (+/- 1,867)
test image_x_portable_bitmap     ... bench:       6,950 ns/iter (+/- 5,136)
test image_x_tga                 ... bench:      20,234 ns/iter (+/- 6,251)
test image_xbm                   ... bench:      40,380 ns/iter (+/- 16,540)
test text_html                   ... bench:      30,258 ns/iter (+/- 9,096)
test text_plain                  ... bench:      32,097 ns/iter (+/- 10,740)

test result: ok. 0 passed; 0 failed; 0 ignored; 18 measured

     Running benches\match_u8.rs (target\release\deps\match_u8-1b76e162075bf416.exe)

running 18 tests
test application_x_7z_compressed ... bench:          79 ns/iter (+/- 15)
test application_x_tar           ... bench:          81 ns/iter (+/- 31)
test application_zip             ... bench:          87 ns/iter (+/- 52)
test audio_flac                  ... bench:          74 ns/iter (+/- 37)
test audio_mpeg                  ... bench:          64 ns/iter (+/- 55)
test audio_ogg                   ... bench:          59 ns/iter (+/- 34)
test audio_opus                  ... bench:          35 ns/iter (+/- 35)
test audio_wav                   ... bench:          76 ns/iter (+/- 38)
test image_bmp                   ... bench:          82 ns/iter (+/- 34)
test image_gif                   ... bench:          71 ns/iter (+/- 25)
test image_png                   ... bench:          77 ns/iter (+/- 24)
test image_tiff                  ... bench:          87 ns/iter (+/- 60)
test image_x_pcx                 ... bench:          94 ns/iter (+/- 127)
test image_x_portable_bitmap     ... bench:         121 ns/iter (+/- 41)
test image_x_tga                 ... bench:          74 ns/iter (+/- 42)
test image_xbm                   ... bench:          32 ns/iter (+/- 28)
test text_html                   ... bench:          88 ns/iter (+/- 47)
test text_plain                  ... bench:          39 ns/iter (+/- 45)

test result: ok. 0 passed; 0 failed; 0 ignored; 18 measured

It's not a substantial improvement, but it's something, and it lets us drop a dependency. Feel free to test on your machine. Keep in mind that individual test performance for from_u8 is dependent in part on whatever order the graph nodes get inserted in.

aehancoc commented 6 months ago

I'm realizing the benchmarks might be easier to read if the benchmarks are ported to Criterion. There's almost certainly a good bit of jitter from the initial lazy load.

mbrubeck commented 6 months ago

Hey, long time no see. Thanks for maintaining this in my absence. (In case you're wondering, my old account was tied to my university email, which I lost access to.)

You’re welcome! And if you’re interested at any point in being the maintainer (or a co-maintainter) of this package again, just let me know. I don’t actually use this library in my other projects anymore, though I’m happy to continue keeping it up to date for others.