[Speed Test] Why is MPMCQueue significantly faster than ProducerConsumerQueue?

[Zheng-Bicheng]

I do a speed test for MPMCQueue and ProducerConsumerQueue. I try to write and read 100,000,000 data in one producer thread and one consumer thread. But the speed of MPMCQueue is significantly faster than the speed of ProducerConsumerQueue. The documentation states that 'ProducerConsumerQueue' is used for single-producer and single-consumer threads, while 'MPMCQueue' is used for multi-producer and multi-consumer threads. Why is MPMCQueue faster even in the case of a single-producer and single-consumer scenario?

[Zheng-Bicheng]

My coding is as follow:

folly::ProducerConsumerQueue<int> BufferProducerConsumerQueue(1000);
void ProducerConsumerQueueProducer() {
    for (int i = 0; i < DATA_NUM; ++i) {
        while (!BufferProducerConsumerQueue.write(i));
    }
}
void ProducerConsumerQueueConsumer() {
    int value;
    for (int i = 0; i < DATA_NUM; ++i) {
        while (!BufferProducerConsumerQueue.read(value));
    }
}
static void TestProducerConsumerQueue(benchmark::State &state){
    for (auto _ : state) {
        std::thread producerThread(ProducerConsumerQueueProducer);  // 创建生产者线程
        std::thread consumerThread(ProducerConsumerQueueConsumer);  // 创建消费者线程
        producerThread.join();  // 等待生产者线程结束
        consumerThread.join();  // 等待消费者线程结束
    }
}
BENCHMARK(TestProducerConsumerQueue)->UseRealTime()->Unit(benchmark::kSecond);

folly::MPMCQueue<int> BufferMPMCQueue(1000);
void MPMCQueueProducer() {
    for (int i = 0; i < DATA_NUM; ++i) {
        while (!BufferMPMCQueue.write(i));
    }
}
void MPMCQueueConsumer() {
    int value;
    for (int i = 0; i < DATA_NUM; ++i) {
        while (!BufferMPMCQueue.read(value));
    }
}
static void TestMPMCQueue(benchmark::State &state){
    for (auto _ : state) {
        std::thread producerThread(MPMCQueueProducer);  // 创建生产者线程
        std::thread consumerThread(MPMCQueueConsumer);  // 创建消费者线程
        producerThread.join();  // 等待生产者线程结束
        consumerThread.join();  // 等待消费者线程结束
    }
}
BENCHMARK(TestMPMCQueue)->UseRealTime()->Unit(benchmark::kSecond);

And the speed results is:

------------------------------------------------------------------------------
Benchmark                                    Time             CPU   Iterations
------------------------------------------------------------------------------
TestProducerConsumerQueue/real_time       8.46 s         0.000 s             1
TestMPMCQueue/real_time                   5.47 s         0.000 s             1