[Ns-developers] [ns-2.33] bug in WirelessPhyExt::sendUp()

[Mathieu Lacage]

On Wed, 2008-06-25 at 16:11 +0300, antoine.trux at nokia.com wrote:

> > the questions are:
> >   a) whether or not a real application will see a difference in
> > simulations (and in the real world)
> 
> This depends on the values of the receiver's real sensitivity and the
> carrier-sense threshold. The larger the difference between these two
> values, the larger the discrepancy between simulation results and the
> real world.

I have never seen a rigorous analysis of a real-world experiment which
shows proof of that. i.e., the difference between the real-world
experiment and the simulation could come from any number of things. Why
would it come from _that_ ? How do you verify it comes from _that_ ?

> >   b) which values of the two thresholds would better modelize a real
> > device, do these values change a lot from one device to the other ?
> 
> I don't know how much the receiver's real sensitivity varies from device
> to device, but the real sensitivity can significantly depart from the
> minimum mandated by the Standard. Suffice it to say that I didn't invent
> the figure of -89 dBm in the above example.

So, where is it coming from ? Part of the reason why 802.11 PHY
simulation models suck so much is that there is little to no litterature
on experiment-backed model design for these PHY layers. So, what we have
is drawn from intuitions from our understanding of the way these devices
are designed, which is quite unlikely to match anything close to a real
world device. Or, if it does, no one wrote about a paper I am aware of
about the success of their intuitions backed by experimental data.

> >   c) whether or not the new model will be really "better" 
> > (that is, its
> > predictions will more accurately reflect real-world experimental
> > measurements)
> 
> Depends on which kind of simulations you are running. See also my answer
> to a) above. Anyway, I see no point in using a false criterion for

I take issue with this "false" adjective. There is nothing "false" about
the current behavior: it is merely an arbitrary model and there are so
many other things in this model which are also completely arbitrary that
pointing out this specific item in isolation without further context is
just meaningless. 

For example, the whole idea that there is something called the "power of
a signal" which is constant over the signal reception is meaningless and
there are many other underlying assumptions about the current model
which are similarly totally meaningless from a physical perspective. So,
is it a "false" model ? Of course it is. But does this model show good
correlation with real-world data ? I don't know and showing proof of
that would go a long way towards really improving the state of the ns-2
(or ns-3) 802.11 PHY models.

> signal reception, even if the simulation results happened to be correct
> for certain parameter values.

Of, course, it depends on which kind of simulation you are running,
there is no doubt about it: it is trivial to devise a simulation which
behaves wildly differently with different parameter values. The
question, though, is whether what you are proposing is really better
from the perpective of modelizing a real world device. Intellectually, I
agree with you that what you suggest is probably a good idea. I have
never seen proof of it though and _that_ was my point. There has been a
recent flurry of activity and papers around that topic, so it might the
case that I have missed something in which case, I would be more than
happy to be pointed out to relevant material :)

Mathieu