-2

I would like to create a few threads in my program where I pass multiple arguments. I found this example where it shows that you can avoid structs and pass multiple arguments to a thread. It worked. Or so I thought. When I looked at the duration of my functions, they took the same amount of time.

I am running my code on Amazon server, so I am not sure if the problem is with the server or my code.

EDIT I added executable code and ran it on my local machine. It doesnt seem to show good results. 

    #include <iostream>
    #include <thread>
    #include <future>
    #include <unistd.h>
    #include "time.h"

    using namespace std;

    void threadCallback(float &number)
    {
        sleep(10);
        number = 10;
    }


    int main()
    {  
        clock_t begin = clock();

        for(int j = 0; j < 2; j++)
        {
            float fitness = 0;
            threadCallback(fitness);

        }

        double duration = double(clock()-begin)/CLOCKS_PER_SEC;
        std::cout << "Original duration \t\t\t" << duration << endl;

        clock_t begin1 = clock();

        float fitness1 = 0;
        std::thread th1(threadCallback, std::ref(fitness1));

        float fitness2 = 0;        
        std::thread th2(threadCallback, std::ref(fitness2));

        th1.join();
        th2.join();

        double duration1 = double(clock()-begin1)/CLOCKS_PER_SEC;
        cout << "My duration \t\t\t" << duration1 << endl;
        return 0;
}

I would appreciate any guidance as I am stuck.

The durations I get are: 0,000095 and 0,000297 respectively.

Gabriele
  • 737
  • 2
  • 8
  • 20
  • 3
    Multi-threading isn't a magic wand that automatically makes code faster. There are costs involved that make code slower. – Paul Evans May 03 '19 at 12:17
  • 3
    Furthermore, until you [edit] your question so that it complies with all the requirements of a [mcve], as explained in stackoverflow.com's [help], that makes it possible to simply cut and paste what you've written in your question, compile it, run it, and reproduce the results you're seeing, nobody will be able to help you. – Sam Varshavchik May 03 '19 at 12:27
  • Followed your advice and added executable code – Gabriele May 03 '19 at 13:49
  • 1
    Whoever marked this as duplicate with the `C` time measurement really didn't understand the problem here (even though it is a valid point). – AlexG May 03 '19 at 14:18

2 Answers2

1

Multithreading is not free. Each time you spawn a thread, there is some initialization overhead. Also, having multiple threads means that the operating system needs to manage them, thus switching from one to another doesn't come for free either.

Your problem (aka the code that needs to run on each thread) is just to small to get any performance benefit. As a general rule of thumb, I wouldn't suggest trying to parallelize any code that takes less than a few milliseconds to run, because depending on your operating system, it might be the time it takes to start another thread.

A second problem you have is with the usage of clock(). It takes into account the time spent running your program, not the elapsed time since the beginning of your program. That's entirely different because a computer never runs a single program at once. To address that, I would suggest looking at QueryPerformanceCounter (that's Windows only). There are probably other functions that are more generic available somewhere...

AlexG
  • 1,091
  • 7
  • 15
  • This is just my toy example. The original code takes a few seconds to run the function, thus why I wanted to parallelize it. – Gabriele May 03 '19 at 14:20
  • 1
    @Gabriele read my update with the usage of `clock`. Also, there are many things to consider when writing parallel code so it might be relevant to write the most accurate code possible. – AlexG May 03 '19 at 14:23
  • Ill mark your answer as the answer. I used chrono high_resolution_clock and it seems to work! Thank you! – Gabriele May 03 '19 at 14:29
0

Run your code on local machine, and check the time. And run same code on amazon server and again check the time.

Usually on cloud it takes some time initially to get started. But once it is executed at least once, then it should not take extra time again.

Check the time by executing your thread function multiple times.

Deepak Agarwal
  • 1
  • 2