Random Number Values quite different based on declaration style

[Mas]

Hi All
I came across a protocol implementation by a researcher. He needs to call a function every second that gets values from uniform distribution in range (0, maxval.size() ) , the value of
maxval, which is a vector, also changes every second and it maximum value is 14. Therefore in every call to function per second the author declares random variable every time like

void getSLotNumber()
{

 UniformVariable x(0, maxval.size() );
 val=x.GetInteger();    
}

1. Is it correct to declare random variable again and again?
   I have used other technique knowing that maxval.size() is maximum 14, by declaring    Ptr<UniformRandomVariable> x initially as class object and then based on the size of current maxval i scale down the retrieved values in desired range (0, currentsize of vector). I did following:  

// declared variable in class
Ptr<UniformRandomVariable> x;
// in constructor initialized variable and assigned range
 x = CreateObject<UniformRandomVariable> ();
double min = 0.0;
double max = 14.0;
x->SetAttribute ("Min", DoubleValue (min));
x->SetAttribute ("Max", DoubleValue (max));

// and called function every second this way
 void getSLotNumber()
   {

    y=x->GetInteger);  
    val=scaleWRT( maxval.size(),y ) // if current vector size is 5 then scales y value in
    // range (0,5) instead of (0,14) using scaleWRT() function created

   }

This way i dont need to redeclare random variable again and again due dynamic size of vairable (vector size)

But the pattern of random variable i got after scaling is quite different from what i get when i used already implemented logic with uniform variable declared every time. Any comments please.

I have also found in ns3 that below two segments of code produce quite differnt patterns

for(int a=0; a<20; a++){
  UniformVariable x(0, 14 );
 val=x.GetInteger();  
}

 //vs
 UniformVariable x(0, 14 );

for(int a=0; a<20; a++){
 val=x.GetInteger();  }

Any Comments on this. For my network simulation both of these produced quite different results. Can any one tell whats the correct way to deal with random numbers?

Thanks

[pdbarnes]

I’m not surprised they give different results: they use different RNG instances. The original implementation draws from a new stream each time, while yours draws from a single stream.

My preference would be to create the RNG once, then set its max value each call to the size of the vector. A single RNG will give repeatable results, independent of what else is happening or created in someone’ model.

Peter