这种异步机制正是 NodeJS 受到欢迎的原因（尽管 NodeJS 不是并行的），以及 Go 作为编程语言的优势所在。

这与 **FastAPI** 的性能水平相同。

您可以同时拥有并行性和异步性，您可以获得比大多数经过测试的 NodeJS 框架更高的性能，并且与 Go 不相上下， Go 是一种更接近于 C 的编译语言（<a href="https://www.techempower.com/benchmarks/#section=data-r17&hw=ph&test=query&l=zijmkf-1" class="external-link" target="_blank">全部归功于 Starlette</a>）。

### 并发比并行好吗？

不！这不是故事的本意。

并发不同于并行。而是在需要大量等待的特定场景下效果更好。因此，在 Web 应用程序开发中，它通常比并行要好得多，但这并不意味着全部。

그리고 **FastAPI**를 사용함으로써 동일한 성능을 낼 수 있습니다.

또한 병렬성과 비동기성을 동시에 사용할 수 있기 때문에, 대부분의 테스트가 완료된 NodeJS 프레임워크보다 더 높은 성능을 얻고 C에 더 가까운 컴파일 언어인 Go와 동등한 성능을 얻을 수 있습니다<a href="https://www.techempower.com/benchmarks/#section=data-r17&hw=ph&test=query&l=zijmkf-1" class="external-link" target="_blank">(모두 Starlette 덕분입니다)</a>.

### 동시성이 병렬성보다 더 나은가?

그렇지 않습니다! 그것이 이야기의 교훈은 아닙니다.

そして、それは**FastAPI**で得られるパフォーマンスと同じレベルです。

また、並列処理と非同期処理を同時に実行できるため、テスト済みのほとんどのNodeJSフレームワークよりも高く、Goと同等のパフォーマンスが得られます。Goは、Cに近いコンパイル言語です <a href="https://www.techempower.com/benchmarks/#section=data-r17&hw=ph&test=query&l=zijmkf-1" class="external-link" target="_blank">(Starletteに感謝します)</a>。

### 並行は並列よりも優れていますか？

いや！それはこの話の教訓ではありません。

	// md5sum of the data to written is required as input for PutObject.

	md5hex := getMD5Hash(textData)
	sha256hex := ""

	// benchmark utility which helps obtain number of allocations and bytes allocated per ops.
	b.ReportAllocs()
	// the actual benchmark for PutObject starts here. Reset the benchmark timer.
	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		// insert the object.

Gunicornと組み合わせることで、非同期でマルチプロセスなサーバーを持つことがきます。

詳細は[デプロイ](deployment/index.md){.internal-link target=_blank}の項目で確認してください。

///

## ベンチマーク と スピード

// This makes it easy to run the TestServer from any of the tests.
// Using this interface, functionalities to be used in tests can be
// made generalized, and can be integrated in benchmarks/unit tests/go check suite tests.
type TestErrHandler interface {
	testing.TB
}

const (
	// ErasureSDStr is the string which is used as notation for Single node ObjectLayer in the unit tests.

			if r != needle {
				rs = append(rs, rune(r))
			}
		}
	}
	// Shuffle the runes so that they are not in descending order.
	// The sort is deterministic since this is used for benchmarks,
	// which need to be repeatable.
	rr := rand.New(rand.NewSource(1))
	rr.Shuffle(len(rs), func(i, j int) {
		rs[i], rs[j] = rs[j], rs[i]
	})
	uchars := string(rs)

	return func(b *testing.B, n int) {

*   Python 3.3+ support via changes to python codebase and ability to specify
    python version via ./configure.

*   Some improvements to GPU performance and memory usage:
    [convnet benchmarks](https://github.com/soumith/convnet-benchmarks/issues/66)
    roughly equivalent with native cudnn v2 performance. Improvements mostly due
    to moving to 32-bit indices, faster shuffling kernels. More improvements to
    come in later releases.

 * limitations under the License.
 */

package com.google.common.collect;

import com.google.caliper.Benchmark;

/**
 * Benchmarking interners.
 *
 * @author Dimitris Andreou
 */
public class InternersBenchmark {
  @Benchmark
  int weakInterner(int reps) {
    Interner<String> interner = Interners.newWeakInterner();
    for (int i = 0; i < reps; i++) {

      factorials[i] = RANDOM_SOURCE.nextInt(100);
    }
  }

  @Benchmark
  long log2(int reps) {
    long tmp = 0;
    for (int i = 0; i < reps; i++) {
      int j = i & ARRAY_MASK;
      tmp += Double.doubleToRawLongBits(DoubleMath.log2(positiveDoubles[j]));
    }
    return tmp;
  }

  @Benchmark
  long factorial(int reps) {
    long tmp = 0;
    for (int i = 0; i < reps; i++) {