An experiment in Go performance

Initial results:

phf@eggbert ~/go/src/pkg/container/hashmap $ time ./example_dict.py 1671705 words

real 0m2.578s user 0m1.820s sys 0m0.140s phf@eggbert ~/go/src/pkg/container/hashmap $ time ./example_map 1671705 words

real 0m1.822s user 0m1.680s sys 0m0.150s phf@eggbert ~/go/src/pkg/container/hashmap $ time ./example_hashmap 1671705 words

real 0m10.216s user 0m9.750s sys 0m0.460s

hashmap.BenchmarkLen 1000000000 3 ns/op hashmap.BenchmarkInsert 500000 4911 ns/op hashmap.BenchmarkRemove 500000 4272 ns/op hashmap.BenchmarkAt 1000000 1510 ns/op hashmap.BenchmarkSet 1000000 1700 ns/op hashmap.BenchmarkSuccessfulLookup 1000000 1442 ns/op hashmap.BenchmarkFailedLookup 5000000 399 ns/op

After replacing container/vector with hashvec:

phf@eggbert ~/go/src/pkg/container/hashmap $ time ./example_hashmap 1671705 words

real 0m5.813s user 0m5.400s sys 0m0.210s

hashmap.BenchmarkLen 1000000000 3 ns/op hashmap.BenchmarkInsert 1000000 1723 ns/op hashmap.BenchmarkRemove 1000000 1007 ns/op hashmap.BenchmarkAt 5000000 501 ns/op hashmap.BenchmarkSet 5000000 687 ns/op hashmap.BenchmarkSuccessfulLookup 5000000 490 ns/op hashmap.BenchmarkFailedLookup 10000000 172 ns/op

Credits

Thanks to Roger Peppe for hashmap_roger.go which uses the builtin map type for half of the data structure.