HTTP parsing benchmarks

This repository holds three different versions of a HTTP parser, written in:

• Rust, with Geal/nom
• Rust, with Marwes/combine
• Rust, with seanmonstar/httparse
• Haskell with bos/attoparsec and GaloisInc/cereal
• C with joyent/http-parser
• C with h2o/picohttpparser

meanings of the various benchmarks:

• one_test is a small HTTP request
• httpparse_example_test is a larger request (long path, complete user agent header and a long cookie)
• small_test is parsing the http-requests.txt file containing 55 HTTP requests
• bigger_test is parsing the bigger.txt file containing 275 HTTP requests

Results

The benchmarks were run on a late 2013 Macbook Pro, quad core 2,3 GHz Intel Core i7. Presented from slowest to fastest

	one_test	small_test	bigger_test	httparse_example_test
attoparsec		241.5 μs/iter (+/-5.7 μs)	1.836 ms/iter (+/- 137 μs)
combine	1,394 ns/iter (+/- 489) = 208 MB/s	89,875 ns/iter (+/- 12,575) = 237 MB/s	453,297 ns/iter (+/- 260,973) = 235 MB/s	2,076 ns/iter (+/- 436) = 338 MB/s
nom	889 ns/iter (+/- 411) = 327 MB/s	59,918 ns/iter (+/- 5,372) = 356 MB/s	333,344 ns/iter (+/- 49,823) = 320 MB/s	1,544 ns/iter (+/- 294) = 455 MB/s
http-parser	606 ns/iter (+/- 367 ns)	56,385 ns/iter (+/- 148 ns)	266254 ns/iter (+/- 37071 ns)	1137 ns/iter (+/- 18 ns)
httparse	213 ns/iter (+/- 39) = 1366 MB/s	15,432 ns/iter (+/- 1,564) = 1385 MB/s	78,121 ns/iter (+/- 12,476) = 1368 MB/s	472 ns/iter (+/- 162) = 1489 MB/s
picohttpparser	170 ns/iter (+/- 29) = 1711 MB/s	13,260 ns/iter (+/- 1,786) = 1612 MB/s	65,415 ns/iter (+/- 5,058) = 1634 MB/s	350 ns/iter (+/- 76) = 2008 MB/s
nom (optimized)	183 ns/iter (+/- 25) = 1590 MB/s	12,253 ns/iter (+/- 2,473) = 1744 MB/s	61,519 ns/iter (+/- 14,660) = 1737 MB/s	308 ns/iter (+/- 48) = 2282 MB/s

(the "nom-optimized" version is an experiment to match httparse and picohttparser APIs, and bring in SIMD based parsing in nom)