[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [ns] M/M/1
Drop tail will be fine, just use a huge buffer (bigger than maximum queue
size in your simulation). Need poisson arrivals (for the first M), and
exponentially distributed packet sizes (for the second M).
This will be approximately correct (and is how Opnet does an M/M/1 example
too), except that in a real M/M/1 queue, departures are a poisson process
with departure rate the same as the arrival rate. In this approximation,
departures will be at your service rate as long as there are bytes in the
buffer.
So, depending on the time resolution you need, this might/might not be
sufficient. What I mean is if you soom in close enough, this approximation
will show a continuous stream of chunks at your service rate, where as M/M/1
would show nothing, followed by a packet sized chunk at one instant, which
could be much different than the service rate.
Clear as mud?
> -----Original Message-----
> From: Debojyoti Dutta [mailto:[email protected]]
> Sent: Monday, March 12, 2001 12:53 PM
> To: Lloyd Wood
> Cc: Radhakrishna Sampigethaya; [email protected]
> Subject: Re: [ns] M/M/1
>
>
> Lloyd is right. I forgot to mention that the source needs to
> be markovian.
> To be precise droptail queueing is M/D/1. I think one can
> change droptail
> Qs and add some randomness to model M/M/1 more accurately.
>
> Debo
>
> On Mon, 12 Mar 2001, Lloyd Wood wrote:
>
> > On Mon, 12 Mar 2001, Debojyoti Dutta wrote:
> >
> > > It must be same / similar to drop tail.
> >
> > ..with arrivals from a single Poisson source, surely? Has to be
> > memoryless.
> >
> > L.
> >
> > > On Wed, 7 Mar 2001, Radhakrishna Sampigethaya wrote:
> > >
> > > > Has anyone tried to implement an M/M/1 queue in ns-2?
> >
> >
> <[email protected]>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>
> >
> >
>