Re: Confidence Interval

[Tarik Alj <aljtarik@ozias.inrs-telecom.uquebec.ca>]

> 
> I am simulating a Multicast protocol, so I am interested to obtain
> numbers of CBR (Call Blocking Rate), but I need to specify the
> confidence interval on my measurements.
> 
> Yesterday I red in the Mailing archives that I have to make the
> calculation for the number of times that I should run the simulation
> (with different seed), in order to get the measurements into a certain
> confidence interval.
> 
> If the above is true, I should run many times the simulation and after
> that change the conditions on the simulation (for example the number of
> active traffic sources) and simulate many times again (?)...

That would be an estimation by replication.
You could also use a batch technique: divide the time (say) of your simulation 
into intervals, estimate over each interval and get an estimation from the 
estimates.

The confidence interval depends on the number and size of batches; or on the 
number (and duration) of replications. 

With ns you should be able to logg whatever happens during a simulation, that's 
what tracefiles are for. From those files you should be able to make estimations 
and deduce your confidence interval.
 
Good luck.


Tarik