vector

During runtime, given a number N (unsigned long int), I have to take 2*N inputs from the user. I have to store them. The task is done and the question is not about how to do it. I was experimenting with the code and how to make it more standard and efficient. So, there are following three options among others :

1. Construct two std::vector<unsigned long int> vectors, to store first and second element respectively of each pair.

2. Construct one std::vector<unsigned long int> vector of length 2*N.

3. Construct one std::vector<std::pair<unsigned long int, unsigned long int> >.

So, I wrote this little program to see the execution speeds.

#include<iostream>                                                                                                                               
#include<vector>                                                                                                                                 
#include<utility>                                                                                                                                
#include<ctime>                                                                                                                                  
using namespace std;                                                                                                                             
int main(int argc, char* argv[]){                                                                                                                
  const unsigned long int len{10};                                                                                                               
  clock_t time1{clock()};                                                                                                                        
  vector<unsigned long int>veca(2*len);                                                                                                          
  //  vector<unsigned long int>vecb(len);                                                                                                        
  cout<<(double)(clock()-time1)/CLOCKS_PER_SEC<<endl;                                                                                            
  time1 = clock();                                                                                                                               
  typedef pair<unsigned long int, unsigned long int> NumberPair;                                                                                 
  vector<NumberPair>vecu(len);                                                                                                                   
  cout<<(double)(clock()-time1)/CLOCKS_PER_SEC<<endl;                                                                                            
  return 1;                                                                                                                                      
}                                                                             

The output was

• 6.8e-05
• 2e-06

If instead of 10, a length of 10^10 is used, then, the output is

• 2e-06
• 1e-06

Thus it is better to go for choice 3 than choice 2.

Similarly by changing the code, it can be shown that choice 3 is better than choice 1.

Moreover the change in execution times, is lesser for large data sizes (execution time halves in case of option 3 !!!), and more for small data sizes (almost 32 to 38 times).

What is the explanation of such a behavior ?