java - 为什么netty FastThreadLocal的基准测试结果没有比ThreadLocal快多少？

Question

我使用netty项目中的benchmark io.netty.microbench.concurrent.FastThreadLocalFastPathBenchmark。

当我直接运行它时，Beachmark的结果显示FastThreadLocal比ThreadLocal快得多。

但是，当我将 get() 方法更改为 set() 方法时，Beachmark 的结果显示 FastThreadLocal 并不比 ThreadLocal 快多少。

这是结果:

Benchmark                                            Mode  Cnt     Score     Error  Units
FastThreadLocalFastPathBenchmark.fastThreadLocal    thrpt   20  1233.992 ± 101.533  ops/s
FastThreadLocalFastPathBenchmark.jdkThreadLocalGet  thrpt   20  1203.562 ±  77.841  ops/s

/*
 * Copyright 2012 The Netty Project
 *
 * The Netty Project licenses this file to you under the Apache License,
 * version 2.0 (the "License"); you may not use this file except in compliance
 * with the License. You may obtain a copy of the License at:
 *
 *   http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
 * License for the specific language governing permissions and limitations
 * under the License.
 */
package io.netty.microbench.concurrent;

import io.netty.microbench.util.AbstractMicrobenchmark;
import io.netty.util.concurrent.FastThreadLocal;
import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.Measurement;
import org.openjdk.jmh.annotations.Threads;

import java.util.Random;

/**
 * This class benchmarks the fast path of FastThreadLocal and the JDK ThreadLocal.
 */
@Threads(4)
@Measurement(iterations = 10, batchSize = 100)
public class FastThreadLocalFastPathBenchmark extends AbstractMicrobenchmark {

    private static final Random rand = new Random();

    @SuppressWarnings("unchecked")
    private static final ThreadLocal<Integer>[] jdkThreadLocals = new ThreadLocal[128];
    @SuppressWarnings("unchecked")
    private static final FastThreadLocal<Integer>[] fastThreadLocals = new FastThreadLocal[jdkThreadLocals.length];

    static {
        for (int i = 0; i < jdkThreadLocals.length; i ++) {
            final int num = rand.nextInt();
            jdkThreadLocals[i] = new ThreadLocal<Integer>() {
                @Override
                protected Integer initialValue() {
                    return num;
                }
            };
            fastThreadLocals[i] = new FastThreadLocal<Integer>() {
                @Override
                protected Integer initialValue() {
                    return num;
                }
            };
        }
    }

    @Benchmark
    public int jdkThreadLocalGet() {
        int result = 0;
        for (ThreadLocal<Integer> i: jdkThreadLocals) {
            i.set(rand.nextInt());
        }
        return result;
    }

    @Benchmark
    public int fastThreadLocal() {
        int result = 0;
        for (FastThreadLocal<Integer> i: fastThreadLocals) {
            i.set(rand.nextInt());
        }
        return result;
    }
}

获取()结果:

Benchmark                                            Mode  Cnt      Score      Error  Units
FastThreadLocalFastPathBenchmark.fastThreadLocal    thrpt   20  51634.074 ± 6936.996  ops/s
FastThreadLocalFastPathBenchmark.jdkThreadLocalGet  thrpt   20  53212.778 ±  660.392  ops/s

Answer 1

嗯，FastThreadLocal 针对 get 进行了优化，而不是针对 set，所以您所看到的一般来说并不奇怪。