| // The trouble with specialized map types... I need a better name. | |
| package main | |
| import "fmt" | |
| type I interface { | |
| M() | |
| } | |
| type IFunc func() |
| package main | |
| import ( | |
| "log" | |
| "os" | |
| mdb "github.com/szferi/gomdb" | |
| ) | |
| func main() { |
Benchmarks were given two seconds to run. Allocations were recorded for each benchmark.
go test -test.benchtime=2s -test.benchmem=true -run=NONE -bench=. > [old,new].bench.txt
Benchmarks are compared with benchcmp
benchcmp old.bench.txt new.bench.txt
| panic: runtime error: index out of range | |
| goroutine 16 [running]: | |
| runtime.panic(0x4b9d80, 0x546cbc) | |
| /usr/local/go/src/pkg/runtime/panic.c:279 +0xf5 | |
| code.google.com/p/snappy-go/snappy.Decode(0x0, 0x0, 0x0, 0x7f713339ceb5, 0xb, 0xb, 0x0, 0x0, 0x0, 0x0, ...) | |
| /home/bryan/src/code.google.com/p/snappy-go/snappy/decode.go:49 +0xb0c | |
| main.main() | |
| /home/bryan/src/github.com/bmatsuo/sandbox/snappy-go-panic/snappy-go-index/main.go:11 +0x9c |
| PASS | |
| BenchmarkReaderManpage 1000000 20223 ns/op 126.09 MB/s | |
| BenchmarkReaderManpage_buffered 1000000 20176 ns/op 126.39 MB/s | |
| BenchmarkReaderManpageNoCopy 1000000 20179 ns/op 126.36 MB/s | |
| BenchmarkReaderJSON 50000 320459 ns/op 76.23 MB/s | |
| BenchmarkReaderJSON_buffered 50000 332016 ns/op 68.00 MB/s | |
| BenchmarkReaderJSONNoCopy 50000 334736 ns/op 67.45 MB/s | |
| BenchmarkReaderRandom 10000 1859415 ns/op 5640.66 MB/s | |
| BenchmarkReaderRandom_buffered 10000 1837270 ns/op 5707.95 MB/s | |
| BenchmarkReaderRandomNoCopy 10000 2782352 ns/op 3769.13 MB/s |
| PASS | |
| BenchmarkReaderManpage 1000000 19525 ns/op 130.60 MB/s | |
| BenchmarkReaderManpage_buffered 1000000 19605 ns/op 130.07 MB/s | |
| BenchmarkReaderManpageNoCopy 1000000 19686 ns/op 129.53 MB/s | |
| BenchmarkReaderJSON 50000 314747 ns/op 77.61 MB/s | |
| BenchmarkReaderJSON_buffered 50000 330732 ns/op 68.27 MB/s | |
| BenchmarkReaderJSONNoCopy 50000 332856 ns/op 67.83 MB/s | |
| BenchmarkReaderRandom 10000 1893232 ns/op 5539.91 MB/s | |
| BenchmarkReaderRandom_buffered 10000 1842477 ns/op 5691.82 MB/s | |
| BenchmarkReaderRandomNoCopy 10000 2765457 ns/op 3792.16 MB/s |
| PASS | |
| BenchmarkReaderManpage 1000000 19398 ns/op 131.45 MB/s | |
| BenchmarkReaderManpage_buffered 1000000 19347 ns/op 131.80 MB/s | |
| BenchmarkReaderManpageNoCopy 1000000 19314 ns/op 132.02 MB/s | |
| BenchmarkReaderJSON 50000 319364 ns/op 76.49 MB/s | |
| BenchmarkReaderJSON_buffered 50000 333882 ns/op 67.62 MB/s | |
| BenchmarkReaderJSONNoCopy 50000 336380 ns/op 67.12 MB/s | |
| BenchmarkReaderRandom 10000 1870011 ns/op 5608.70 MB/s | |
| BenchmarkReaderRandom_buffered 10000 1843386 ns/op 5689.01 MB/s | |
| BenchmarkReaderRandomNoCopy 10000 2748211 ns/op 3815.95 MB/s |
| # benchcmp bench.new.txt bench.newinline.txt | |
| benchmark old ns/op new ns/op delta | |
| BenchmarkReaderManpage 19525 19398 -0.65% | |
| BenchmarkReaderManpage_buffered 19605 19347 -1.32% | |
| BenchmarkReaderManpageNoCopy 19686 19314 -1.89% | |
| BenchmarkReaderJSON 314747 319364 +1.47% | |
| BenchmarkReaderJSON_buffered 330732 333882 +0.95% | |
| BenchmarkReaderJSONNoCopy 332856 336380 +1.06% | |
| BenchmarkReaderRandom 1893232 1870011 -1.23% |
#Benchmark results created with go test
Test data was generated on a 4-core i7 (16GB ram) Lenovo ThinkPad running Ubuntu 14.04.
Each benchmark was run for 10 seconds (10x the default) to reduce noise seen when the benchmarks ran for 1 second.
go test -run=NONE -bench=Reader -benchtime=10s
I've included the raw test results for each of the master branch, the original pull request, and the pull request modified to avoid heap allocation by inlining the copyWriter struct literal. For the latter two I've included benchcmp output comparing them to the master branch. There is also a comparison between the pull request and it's modified form.
The build script in this gist occasionally crashes compiling even the simplest of Go programs (main.go) with the traceback in stderr.log. The strace log corresponding to the crash is in strace.txt.
The build script needs to be invoked with the -c flag because the behavior does not always exhibit.