Java – synchronized vs readwritelock performance

I try to prove that synchronized slows down when there are many readers and only a few authors Somehow, I proved the opposite

RW example, execution time 313 MS:

package zad3readWriteLockPerformance;

import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReadWriteLock;
import java.util.concurrent.locks.reentrantreadwritelock;

public class Main {
    public static long start,end;

    public static void main(String[] args) {
        Runtime.getRuntime().addShutdownHook(new Thread(() -> {
            end = System.currentTimeMillis();
            System.out.println("Time of execution " + (end - start) + " ms");
        }));
        start = System.currentTimeMillis();
        final int NUMBER_OF_THREADS = 1000;
        ThreadSafeArrayList<Integer> threadSafeArrayList = new ThreadSafeArrayList<>();
        ArrayList<Thread> consumerThreadList = new ArrayList<Thread>();
        for (int i = 0; i < NUMBER_OF_THREADS; i++) {
            Thread t = new Thread(new Consumer(threadSafeArrayList));
            consumerThreadList.add(t);
            t.start();
        }

        ArrayList<Thread> producerThreadList = new ArrayList<Thread>();
        for (int i = 0; i < NUMBER_OF_THREADS/10; i++) {
            Thread t = new Thread(new Producer(threadSafeArrayList));
            producerThreadList.add(t);
            t.start();

        }



        //  System.out.println("Printing the First Element : " + threadSafeArrayList.get(1));

    }

}
class Consumer implements Runnable {
    public final static int NUMBER_OF_OPERATIONS = 100;
    ThreadSafeArrayList<Integer> threadSafeArrayList;

    public Consumer(ThreadSafeArrayList<Integer> threadSafeArrayList) {
        this.threadSafeArrayList = threadSafeArrayList;
    }

    @Override
    public void run() {
        for (int j = 0; j < NUMBER_OF_OPERATIONS; j++) {
            Integer obtainedElement = threadSafeArrayList.getRandomElement();
        }
    }

}
class Producer implements Runnable {
    public final static int NUMBER_OF_OPERATIONS = 100;
    ThreadSafeArrayList<Integer> threadSafeArrayList;

    public Producer(ThreadSafeArrayList<Integer> threadSafeArrayList) {
        this.threadSafeArrayList = threadSafeArrayList;
    }

    @Override
    public void run() {
        for (int j = 0; j < NUMBER_OF_OPERATIONS; j++) {
            threadSafeArrayList.add((int) (Math.random() * 1000));
        }
    }

}

class ThreadSafeArrayList<E> {
    private final ReadWriteLock readWriteLock = new reentrantreadwritelock();

    private final Lock readLock = readWriteLock.readLock();

    private final Lock writeLock = readWriteLock.writeLock();

    private final List<E> list = new ArrayList<>();

    public void add(E o) {
        writeLock.lock();
        try {
            list.add(o);
            //System.out.println("Adding element by thread" + Thread.currentThread().getName());
        } finally {
            writeLock.unlock();
        }
    }

    public E getRandomElement() {
        readLock.lock();
        try {
            //System.out.println("Printing elements by thread" + Thread.currentThread().getName());
            if (size() == 0) {
                return null;
            }
            return list.get((int) (Math.random() * size()));
        } finally {
            readLock.unlock();
        }
    }

    public int size() {
        return list.size();
    }

}

For the synchronization example, the execution time is only 241ms:

package zad3readWriteLockPerformanceZMIENONENENASYNCHRO;

import java.util.ArrayList;
import java.util.Collections;
import java.util.List;

public class Main {
    public static long start,end;

    public static void main(String[] args) {
        Runtime.getRuntime().addShutdownHook(new Thread(() -> {
            end = System.currentTimeMillis();
            System.out.println("Time of execution " + (end - start) + " ms");
        }));
        start = System.currentTimeMillis();
        final int NUMBER_OF_THREADS = 1000;
        List<Integer> list = Collections.synchronizedList(new ArrayList<Integer>());
        ArrayList<Thread> consumerThreadList = new ArrayList<Thread>();
        for (int i = 0; i < NUMBER_OF_THREADS; i++) {
            Thread t = new Thread(new Consumer(list));
            consumerThreadList.add(t);
            t.start();
        }

        ArrayList<Thread> producerThreadList = new ArrayList<Thread>();
        for (int i = 0; i < NUMBER_OF_THREADS / 10; i++) {
            Thread t = new Thread(new Producer(list));
            producerThreadList.add(t);
            t.start();
        }

        //  System.out.println("Printing the First Element : " + threadSafeArrayList.get(1));

    }

}

class Consumer implements Runnable {
    public final static int NUMBER_OF_OPERATIONS = 100;
    List<Integer> list;

    public Consumer(List<Integer> list) {
        this.list = list;
    }

    @Override
    public void run() {
        for (int j = 0; j < NUMBER_OF_OPERATIONS; j++) {
            if (list.size() > 0)
                list.get((int) (Math.random() * list.size()));
        }
    }

}

class Producer implements Runnable {
    public final static int NUMBER_OF_OPERATIONS = 100;
    List<Integer> threadSafeArrayList;

    public Producer(List<Integer> threadSafeArrayList) {
        this.threadSafeArrayList = threadSafeArrayList;
    }

    @Override
    public void run() {
        for (int j = 0; j < NUMBER_OF_OPERATIONS; j++) {
            threadSafeArrayList.add((int) (Math.random() * 1000));
        }
    }

}

Why is synchronous collection faster when I have ten times more readers than authors How do I show the progress of RW locks I have read in many articles?

Solution

The actual cost of acquiring readwritelock is usually much slower than that of acquiring simple mutexes Javadoc entry of readwritelock:

Therefore, the fact that your thread performs very simple operations may mean that performance depends largely on the time actually spent acquiring locks

Another problem with your benchmark is math Random is synchronous Start with Javadoc:

Therefore, even if your concurrent readers don't block each other when getting readwritelock, they may still compete in math Random, thus defeating some advantages of using readwritelock You can improve this by using threadlocalrandom The usage of this class should normally take the form threadlocalrandom current(). Nextx (...) (where x is int, long, etc.)

In addition, as assyrias points out, naive Java benchmarks without considering JIT compilation and other runtime quirks are unreliable You should use benchmarks such as Java micro benchmarking harness (jmh)