Mailinglist Archive: zypp-devel (83 mails)

[Josef Reidinger <jreidinger@xxxxxxx>]

Duncan Mac-Vicar Prett wrote:
> Stanislav Visnovsky wrote:
> > IIRC, Duncan has a half-finished patch for download in threaded way.
> >
> > Stano
> >   
> (only internal)
> http://w3.suse.de/~dmacvicar/libzypp/0001-first-attempt-to-do-paralell-downloads.patch 
>
>
>
> It does a nice crash :-)

I see that duncan use boost::thread and because I never use it I write 
little performance test for boost::thread if pthreads isn't better. And 
result is that it is almost same (different is statistical error).

boost threads:
user time: 4032
system time: 24

pthreads:
user time: 4076
system time: 14

source code attached.

Pepa
#include <iostream>
#include <sys/param.h>
#include <sys/times.h>
#include <sys/types.h>
#include <boost/thread/thread.hpp>
#include <pthread.h>

#define LOOP 1000

struct tms t,u;

using namespace std;

struct test{
  void operator()() {
    int j = 1;
    for (int i = 0;i<1000000;i++)
      j += i;
  }
};

void* testp(void*){
  test t;
  t();
  return NULL;
}

int main(){

  cout << "boost threads: " <<endl;
  times(&t);
  for (int i=0;i<LOOP;i++)
  {
    test t;
    boost::thread* threads[10];
    for (int i = 0;i<10;++i){
      threads[i] = new boost::thread(t);
    }
    for (int i = 0;i<10;++i){
      threads[i]->join();
      delete threads[i];
    }
  }
  times(&u);
  cout << "user time: " << u.tms_utime-t.tms_utime << endl;
  cout << "system time: " << u.tms_stime-t.tms_stime << endl << endl;

  cout << "pthreads:" <<endl;
  times(&t);
  for (int i=0;i<LOOP;i++)
  {
    pthread_t threads[10];
    test t;
    for (int i = 0;i<10;++i){
      pthread_create(threads+i,NULL,testp,NULL);
    }
    for (int i = 0;i<10;++i){
      pthread_join(threads[i],NULL);
    }
  }
  times(&u);
  cout << "user time: " << u.tms_utime-t.tms_utime << endl;
  cout << "system time: " << u.tms_stime-t.tms_stime << endl;
  
  return 0;
}