[Ns-developers] ns3: random numbers
mweigle at cs.odu.edu
Mon Mar 12 10:58:54 PDT 2007
On Mar 12, 2007, at 12:49 PM, mathieu lacage wrote:
> On Mon, 2007-03-12 at 12:30 -0400, Michele Weigle wrote:
>> To generate multiple independent replications of a simulation (for
>> example, for computing the mean and confidence interval of some
>> metric), you should set the runNumber global variable along with the
>> global seed. The global seed should be the same for all runs of the
>> simulation, and runNumber should be incremented for each run of the
>> simulation. Behind the scenes, the RNG will seed the RandomVariables
>> such that not only are all RandomVariables within a simulation run
>> independent, but that all RandomVariables over all simulation runs
>> are independent.
> The following question is probably naive but I will shoot anyway.
> you avoid using a runNumber variable and keep on seeding all new
> from the same initial seeding stream during the subsequent runs ? Or
> would this have obviously bad properties from the point of view of
> randomness ?
So, instead of using substreams, just take the last seed used from
the current run and use it as the global seed for the next run? I
don't see any problem with that. This RNG has 1.8x10^19 independent
streams, so you shouldn't ever run out. We'd just have to include a
function that returns the next seed to be used so that you can save
it and use it as the global seed in your next run.
FYI -- for a diagram (from L'Ecuyer) of streams vs. substreams, see
More information about the Ns-developers