Re: [ns] Random Variables in ns (2)

["Michele C. Weigle" <clark@cs.unc.edu>]

On Thu, 11 Oct 2001, Miguel Angel Alvarez wrote:
> I am a student of Telecommunications Engineering and I am using ns for
> simulating internet traffic sources. I need to implement a Weibull
> distribution and a Gamma distribution random variable generators. I have
> been researching how ns implements its random variables (exponential,
> pareto, lognormal,...) but I have some doubts about this. Please, I would
> like to know:
>
> Which files are involved in the implementation of random variables in ns?
> I think that they are: random.cc  .h, ranvar.cc  .h, rng.cc  .h
> I tried to develope a Weibull distribution but the constructor failed, so I
> think that it's necessary to modify some other files, so I would be very
> gratefull if someone told me some instructions for developing new random
> variables into ns source.

Hi,

  I've successfully added a Weibull random variable.  I added code
to ranvar.cc, ranvar.h, random.h, and rng.h. Here's what I did:

1) Added the following code to ranvar.cc.  This sets up
RandomVariable/Weibull that you can call from a TCL script.

static class WeibullRandomVariableClass : public TclClass {
 public:
        WeibullRandomVariableClass() : TclClass("RandomVariable/Weibull") {}
        TclObject* create(int argc, const char*const* argv) {
                if (argc >= 6) {
                        return (new WeibullRandomVariable (
				(double) atof(argv[4]),
				(double) atof(argv[5])));
                }
                else {
                        return(new WeibullRandomVariable());
                }
        }
} class_weibullranvar;

WeibullRandomVariable::WeibullRandomVariable()
{
        bind("shape_", &shape_);
        bind("scale_", &scale_);
}

double WeibullRandomVariable::value()
{
        return(rng_->rweibull(scale_, shape_));
}

2) Added the defs for these fns in ranvar.h.

class WeibullRandomVariable : public RandomVariable {
public:
        virtual double value();
        WeibullRandomVariable();
        WeibullRandomVariable(double shape, double scale);
        double* shapep() { return &shape_; };
        double* scalep() { return &scale_; };
        double shape()   { return shape_; };
        double scale()   { return scale_; };
        void setshape(double d)  { shape_ = d; };
        void setscale(double d)  { scale_ = d; };
private:
        double shape_;
        double scale_;
};

3) Added the following line to rng.h.  Here's the math.

       inline double rweibull(double shape, double scale) {
                return (pow (-log (uniform()), 1/shape) * scale);
        }

4) Added the following line to random.h.  I'm not positive this line is
needed, but I was following the format of lognormal() and it was included
here, too.

static double rweibull(double shape, double scale) {return
	rng()->rweibull(shape, scale);}


  Implementing Gamma should probably follow the same form.  I don't know
anything about the empirical distribution implementation, though.

Good Luck,
Michele