[Ns-developers] Numerical values of the attributes inYansWifiPhy::GetTypeId

[Ruben Merz]

Hagen Paul Pfeifer wrote:
> On Tue, 24 Feb 2009 19:39:42 +0100, Ruben Merz
> 
> <ruben at net.t-labs.tu-berlin.de> wrote:
> 
>> Hagen,
> 
> 
>> It looks like my question was not clear enough. I know what those
> 
>> parameters represent. My question is whether any of them had been
> 
>> validated.
> 
> 
> 
> Ok, I misunderstood your question. You shake on the fundamental skeleton
> 
> of a network simulator and your question is absolutely right! But I am
> 
> sure that nobody can guarantee a real-world behavior - this is *the* cost
> 
> of a
> 
> simulator.
> 
> 
> 
> The best that someone can do is to validate simulator results with other
> 
> simulators/emulator/whatever. You dig into the PHY - but the PHY and all
> 
> surrounding algorithms like fading, shadowing, BER/SINR based RX are so
> 
> complex and embossed by noise, nobody can provide a perfect enough model.
> 
> 
> 
> For example: sometimes I must deal with fading (especially doppler
> 
> shifts) and frequency issues - the resulting variances are partly
> 
> unexpected high!
> 
> Take the Nakagami model: how many people that use ns-3 knows all knobs?
> 
> How many people know about impact of a different frequency, et cetera.
> 
> 
> 
> At the end: the only thing you can make, is to make sure that the
> 
> underlying
> 
> model is not to incorrect, compare with other simulation, emulation and
> 
> real world environments (but the later is often misleading) and keep the
> 
> models as simple as possible. Often people should focus on
> 
> comprehensibility.
> 
> Look at the thousands of papers - nobody knows what actual had simulated.


I think that the real problem is not whether the model is simple or not,
or even comprehensible. I think what's missing currently is that (for
the PHY) first, we have no clear idea of what's really wrong or right
for PHY modeling. You can't have detailed models of everything, but if
you model something you should know clearly what the assumptions are and
what you're loosing in terms of simulation accuracy. Second, we lack
validation of the models.

Best,
Ruben