go map 的内存高效实现？

Question

我的用例是通过网络传输一组成员（整数），所以我们采用增量编码，在接收端我们解码并将整个列表作为映射，map[string]struct{} 复杂度为 O(1)用于会员检查。

我面临的问题是，对于 200 万个整数，成员的实际大小仅为 15MB，但堆中映射的大小为 100+MB。似乎 Go 的实际地图实现不适合大型地图。由于它是一个客户端 SDK，我不想对可用内存产生太大影响，并且可能有多个这样的组需要在内存中保存很长时间——大约 1 周。

为此，Go 中是否有更好的替代 DS？

type void struct{}

func ToMap(v []int64) map[string]void {

 out := map[string]void{}

 for _, i := range v {

   out[strconv.Itoa(int(i))] = void{}

 }

 return out

}

Answer 1

这是一种更节省内存的地图形式：type void struct{}func ToMap(v []int64) map[int64]void {&nbsp; &nbsp; m := make(map[int64]void, len(v))&nbsp; &nbsp; for _, i := range v {&nbsp; &nbsp; &nbsp; &nbsp; m[i] = void{}&nbsp; &nbsp; }&nbsp; &nbsp; return m}Go 映射针对整数键进行了优化。通过给出确切的地图大小作为提示来优化地图分配。Astring有一个隐式指针，它会使垃圾收集器 (gc) 每次扫描时都遵循该指针。这是 200 万个伪随机整数的 Go 基准测试：package mainimport (&nbsp; &nbsp; "math/rand"&nbsp; &nbsp; "strconv"&nbsp; &nbsp; "testing")type void struct{}func ToMap1(v []int64) map[string]void {&nbsp; &nbsp; out := map[string]void{}&nbsp; &nbsp; for _, i := range v {&nbsp; &nbsp; &nbsp; &nbsp; out[strconv.Itoa(int(i))] = void{}&nbsp; &nbsp; }&nbsp; &nbsp; return out}func ToMap2(v []int64) map[int64]void {&nbsp; &nbsp; m := make(map[int64]void, len(v))&nbsp; &nbsp; for _, i := range v {&nbsp; &nbsp; &nbsp; &nbsp; m[i] = void{}&nbsp; &nbsp; }&nbsp; &nbsp; return m}var benchmarkV = func() []int64 {&nbsp; &nbsp; v := make([]int64, 2000000)&nbsp; &nbsp; for i := range v {&nbsp; &nbsp; &nbsp; &nbsp; v[i] = rand.Int63()&nbsp; &nbsp; }&nbsp; &nbsp; return v}()func BenchmarkToMap1(b *testing.B) {&nbsp; &nbsp; b.ReportAllocs()&nbsp; &nbsp; b.ResetTimer()&nbsp; &nbsp; for N := 0; N < b.N; N++ {&nbsp; &nbsp; &nbsp; &nbsp; ToMap1(benchmarkV)&nbsp; &nbsp; }}func BenchmarkToMap2(b *testing.B) {&nbsp; &nbsp; b.ReportAllocs()&nbsp; &nbsp; b.ResetTimer()&nbsp; &nbsp; for N := 0; N < b.N; N++ {&nbsp; &nbsp; &nbsp; &nbsp; ToMap2(benchmarkV)&nbsp; &nbsp; }}输出：$ go test tomap_test.go -bench=.BenchmarkToMap1-4&nbsp; &nbsp; &nbsp;2&nbsp; 973358894 ns/op&nbsp; &nbsp; 235475280 B/op&nbsp; &nbsp; 2076779 allocs/opBenchmarkToMap2-4&nbsp; &nbsp; 10&nbsp; 188489170 ns/op&nbsp; &nbsp; &nbsp;44852584 B/op&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;23 allocs/op$&nbsp;