I've released on github three Go packages that present the same set of functions as crypto/md5, crypto/sha1 and crypto/rc4 but that use the OpenSSL libcrypto functions instead of the native Go implementation.
Testing by sending 4.4GB of data through these three functions yields the following results (the test document was the King James Bible concatenated 1,000 times and sent to MD5, SHA1 and RC4 in 4.4MB chunks). All tests were done using go version devel +7dc8d66efb6d Mon Jan 21 10:53:39 2013 +1100 linux/amd64.
These packages provide the same interface as the native Go implementations and should be drop in replacements. libcrypto must be available. The code is available in my go-openssl repository.
PS As this is the first time I've written any cgo code, I'd be happy to hear from more experienced Go programmers about improvements to my wrappers.
PPS As a comparison here is the output of the Go benchmark tests on the native Go implementations of MD5 and SHA1
Testing by sending 4.4GB of data through these three functions yields the following results (the test document was the King James Bible concatenated 1,000 times and sent to MD5, SHA1 and RC4 in 4.4MB chunks). All tests were done using go version devel +7dc8d66efb6d Mon Jan 21 10:53:39 2013 +1100 linux/amd64.
MD5 | SHA1 | RC4 | |
---|---|---|---|
Native Go | 404 MB/s | 123 MB/s | 111 MB/s |
Via OpenSSL | 607 MB/s | 636 MB/s | 744 MB/s |
Speedup | 1.5x | 5.2x | 6.7x |
These packages provide the same interface as the native Go implementations and should be drop in replacements. libcrypto must be available. The code is available in my go-openssl repository.
PS As this is the first time I've written any cgo code, I'd be happy to hear from more experienced Go programmers about improvements to my wrappers.
PPS As a comparison here is the output of the Go benchmark tests on the native Go implementations of MD5 and SHA1
% go test -bench=".*" crypto/md5 PASS BenchmarkHash8Bytes 5000000 477 ns/op 16.74 MB/s BenchmarkHash1K 1000000 2882 ns/op 355.26 MB/s BenchmarkHash8K 100000 20061 ns/op 408.34 MB/s BenchmarkHash8BytesUnaligned 5000000 477 ns/op 16.76 MB/s BenchmarkHash1KUnaligned 1000000 2875 ns/op 356.15 MB/s BenchmarkHash8KUnaligned 100000 20147 ns/op 406.60 MB/s ok crypto/md5 15.997s % go test -bench=".*" crypto/sha1 PASS BenchmarkHash8Bytes 2000000 799 ns/op 10.00 MB/s BenchmarkHash1K 200000 9098 ns/op 112.54 MB/s BenchmarkHash8K 50000 67341 ns/op 121.65 MB/s ok crypto/sha 18.364sAnd here's the same with the OpenSSL wrapped versions:
% go test -bench=".*" md5 PASS BenchmarkHash8Bytes 5000000 611 ns/op 13.09 MB/s BenchmarkHash1K 1000000 2272 ns/op 450.54 MB/s BenchmarkHash8K 100000 13816 ns/op 592.92 MB/s BenchmarkHash8BytesUnaligned 5000000 616 ns/op 12.97 MB/s BenchmarkHash1KUnaligned 1000000 2282 ns/op 448.59 MB/s BenchmarkHash8KUnaligned 100000 14089 ns/op 581.44 MB/s ok md5 15.049s % go test -bench=".*" sha1 PASS BenchmarkHash8Bytes 5000000 625 ns/op 12.78 MB/s BenchmarkHash1K 1000000 2249 ns/op 455.31 MB/s BenchmarkHash8K 100000 13507 ns/op 606.46 MB/s ok sha1 7.525s
No comments:
Post a Comment