How to run boost::threads on many processors?

问题:

I work with a virtual machine computer cluster with many amd64 processors and Debian Squeeze. Previously, I've successfully executed shell scripts in parallel on it (with GNU Parallel). Now, I'd like to use boost::threads. I run this program:

#include <boost/thread.hpp>

using namespace std;
boost::thread_group g;

void foo()
{
    for(int i = 0; i < 1000000000; ++i) 
        for(int i = 0; i < 1000000000; ++i);
}

int main(int argc, char* argv[])
{
    g.add_thread(new boost::thread(foo));
    g.add_thread(new boost::thread(foo));
    g.add_thread(new boost::thread(foo));

    g.join_all();
}

All these threads run on a single processor, that is used in 300% (according to top command). How to make these three threads to run on three separate processors? Is it possible with boost::threads?

Note: This Multiprocessor Boost::Thread? All threads running on one processor, despite the title, is about multicore system whereas mine is truly about multiprocessor system.

回答1:

It's running correctly. You're spawning three threads and all three are running concurrently. The reported CPU usage in multi-threaded applications is the sum of the CPU usage of all the threads. You have three threads that each use 100%, therefore you have 300% usage.

Multiple cores and multiple CPU sockets appear the same in nearly all threading libraries. You have to go out of your way to tell the difference, eg. libNUMA.