[e2e] end2end-interest Digest, Vol 19, Issue 11

[Detlef Bosau]

This directly relates to my own research problem and I´m glad about this 
question.

Our whole discussion yesterday was about how we simulate, how we prevent 
botchers from altering the ns, and we can continue it to why we use the 
ns, why not opnet, why not glomosim etc. etc. etc.

At least, I prefer an open source program because I can read the source 
and see, what is simulated.

_Your_ question is: Is there an evidence that there is some, if loose, 
correlation or similarity bewtween the simulation in the c-code and reality.

And it does not matter here whether we do simulation or emulation or 
what *mulation ever. The core is a simulation and we must make sure that 
this matches reality.


Let me illustrate my own question. Let´s draw a network.

FH-------(the Internet)------PEP/PS----(some mobile wireless wide area 
network)----Mobile.

What I would like to know is:

When the PEP does connection splitting (I-TCP, Bakre and Badrinath, I 
think 1994 or so), should he do dumb spoofing, i.e. ACK TCP packets as 
they arrive, or should he do Path Tail Emulation?

So, we have basically two issues here.

First: How does I-TCP control the sender? I does so via TCP flow 
control. In the paper by Bakre and Badrinath, I-TCP was evaluated with a 
prototypical implementation. Just to say that.

The NS does not implement TCP flow control. That´s what my AWND question 
was about.

Second: How do we simulate the mobile network? And _that´s_ the 1 
million dollar question here, because this makes the difference whether 
a reviewer will recognize is own idea of mobile networks  here 
("Brillant!, Excellent work! Strong Accept! Best Paper Award!"), whether 
he does not believe in simulation anyway ("Full of Shit!", "Awkward", 
"Strong Reject") or whether there is something in between.

Let me sharpen this: _This_ part of the simulation / evaluation is the 
very point, where results can be faked. That caused my aggravaion about 
the AP-TCP paper from Lindemann´s department, which I simply don´t buy.

And even more: One can easily fool oneself here. And I often did.

Unfortunately, and that´s why I have to use a simulation here, I have 
neither and UMTS testbed not an industrial partner which lets me test 
something in his testbed, of course under non disclosure and I cannot 
access the code at the PEP ;-)

In addition: What´s reality? Is the one irreproducible observation in 
one single experiment more convincing than the one simulation? I don´t 
think so.

So _either_ we have _reproducible_ experiments (which is not easy in 
wireless networks) or we have convincing simulations (which is not easy 
as well).

So, the ideal way is a proper rationale. Is there a convincing 
analytical model for wireless channels? Is it moreover practically 
useful? I mean: For _practical_ questions? (Pollaczeck-Chinchin´s 
Theorem....;-)) And: Is this validated in experiments?

O.k. Now we end up in the theory, how we should validate a theory. And 
because I´m a computer scientist and neither Plato nor Sir Karl Popper I 
won´t continue this. I must leave this to experts.

For me, there is no alternative to use a simulation which is based upon 
a proper idea of the behaviour of the path from PEP to the mobile.

Honestly, I must admit: I´m totally helpless there. I did not yet see 
even _one_ _single_ publication with a complete description what even 
_happens_ here and how this affects a packet flow from PEP to MH.

So, the AWND problem is simple programming here.

But the model for the mobile network will make the difference whether my 
work is a waste of time or whether it´s useful.

And I would never rely on commercial programs here, where I can´t see 
the source and have no idea what happens.

Perhaps you could arrange that I can make real measurements in a testbed 
of Nokia? This would be much more convincing than any simulation.

Once again: _This_ is the trapdoor where "results" are often faked.

Perhaps, I personally will start with a description, hat happens on the 
wireless channel and what affects the packet transportation. Only to get 
the "big picture", what happens in general and what must be studied.

However, we need data for this. E.g. roaming. How long does it typially 
take? What´s the typcial duration of a short time disconnection here?
E.g.: Delay spikes. Are they a relevant problem? Thierry Klein gave one 
reason for dely spikes: The algorithm used for opportunistic scheduling. 
Can this be improved? Thierry argued in this direction.

What about Voice, QoS / Packet, Best Effort scheduling?

What about Interleaving?

All these are generic techniques, independent from GPRS or UMTS, which 
may affect the transport latency from PEP to the Mobile.

I will have to study this in simulations. But first of all, I have to 
understand what I have to simulate ....

And it must be similar to reality .....

Detlef

Yogesh Prem Swami wrote:
> I have a somewhat general question about simulations. My question is
> that is there any scientific reason why simulations should be able to
> predict the behavior of a real packet transmission phenomena? Unless you
> make the assumption that packet transmission/interaction is a
> non-chaotic phenomenon (chaotic, as used in physics), there is no reason
> to believe why a simulation would be able to model real world events.
> 
> In other words, how did the networking community come to the conclusion
> that the error between a simulation results and real world packet
> transmission would be bounded, if someone ran the simulations long
> enough? Also, stability of mean and variance only signifies that the
> system (simulator) has a saddle point, but nothing more. (Although, I do
> agree, that this is the very least a researcher can do.)
> 
> To give an analogy, take weather prediction for example. Accurately
> predicting weather is often hard, because the "weather" is fundamentally
> a chaotic event that can not be easily simulated. If packet
> transmissions are also like weather, then there is no reason why
> simulations and real world implementation of protocols will have any
> similarity in their behavior.
> 
> Unless someone shows me a proof that packet transmissions fall within
> the non-chaotic region, I will have a hard time accepting or advocating
> the use of NS-2 or any other simulation tool.
> 
> Thanks
> Yogesh
> 
> ext S. Keshav wrote:
> 


-- 
Detlef Bosau
Galileistrasse 30
70565 Stuttgart
Mail: detlef.bosau at web.de
Web: http://www.detlef-bosau.de
Mobile: +49 172 681 9937