Hi there,

some thoughts on setting simulation paramerters. I hope they will be 
helpful for you even though my answer is not a direct answer to your 
questions.

1. you want to compare "performance". Therefore, it is necessary that 
you spend some thought on what you define as "performance". System 
throughput? Single user throughput? HTTP response time for single users? 
Average/min/max of these? Number of route changes? Time for route 
discovery? Number of hops for a given route? Degree of connectednes in 
the network? etc.

Basically, "performance" can only be measured by performance indicators 
and many of the above mentioned things trade-off on each other. Eg. if 
you change something in the system and it improves eg. HTTP response 
times for single users this change might degrade system response time at 
the same time. So you need to define exactly what is important for you 
and how far you want to trade-off. Eg. if you can improve HTTP response 
times by x% you are willing to accept a degradation of system throughput 
of y%. Hopefully you find a method so that x > y ;-)


2. when you have defined your performance goal you can start to think 
about how to test for it. Eg. if you know that one of your protocols is 
better for dense networks and one is better in sparse networks then you 
should vary the density of the network and eg. check whether oen 
protocol covers a wider variety of situations, if there is any overlap 
and how they compare in the overlap region, etc. For this you should 
check existing literature to know the features of the protocols 
beforehand pretty well.

These two steps are really important, otherwise you will waste lots of 
time with mindless simulation.


3. Examples:

Are you sure why you chose 10 source/destination pairs? Should you vary 
this number? How do you expect that this number will affect the two 
protocols you want to compare?

Same goes for the number of 50 nodes in total.

Why is the topology 1500x300? btw: 1500x300 what? meters? feet?

Is 50 nodes on a 1500x300 (lets assume m^2) topology considered dense or 
sparse for the protocols you want to investigate? What does the 
literature suggest?

The topo is only 300m wide but 1500m long. Have you thought about how 
you are going to generate the positions of the nodes (in theory)? I 
expect that you want to distribute them "uniformly". Are you just 
generating x and y positions with x~uniform and y~uniform? This would 
result in a higher density of the nodes in one direction over the other. 
Is that really what you want for "uniform"?


4. some practical hints:

I usually write generation scripts in tcl with which I generate the 
position and movement files myself before the simulation itself. I then 
save the position and movement files and use them for each of the 
simulations where I want to change eg. just the traffic but not the 
movement pattern. In this way I get reproducability. If you just 
generate the movement on-the-fly and don't save it to a file you can 
never re-run a simulation.

eg. I make a script called generator.tcl which takes eg. number of nodes 
(say, 50) and topology (say, 1500x300) size as input and generates (with 
print to file) a file, say n-50-t-1500-300.scen . This is just a text 
file which contains tcl commands to set the x and y positions of the 
nodes and the tcl commands to set up the topology. In the same way I can 
generate a matching movement file which contains movement commands in 
tcl. In this way, I can try several different movement patterns on the 
same initial topology. Later on I source these files in my main 
simulation script. I usually run several runs of the simulation driven 
by a shell script overnight or over the weekend. The shell script can 
give parameters to the simulation script and depending on these 
parameters I can load the corresponding scenario and movement files. Not 
only that, the shell script can also make sure that any missing scenario 
files are generated before the simulation is run. After the simulation 
runs I usually automatically do some pre-processing of the trace files 
to save disk space. And, of course, I can run many simulations with 
different settings for the random generator.

So the next morning I can have the results for say, 20, 50, and 100 
nodes in the same topology, all averaged over 30 simulation runs each.


good luck,

Martina Umlauft


vitz399 schrieb:
> 
> Hello everybody.
> 
> I have a write a report about the performance comparison between AODV and
> DSR. Now, I have to setup a simulation environment for both AODV and DSR
> protocols. But I have no idea how could I configure the parameters for
> simulations. Here is the initial parameters for the simulations:
> 
> 10 source/destination pairs for a
> 1500x300 network of
> 50 nodes.
> 
> The source should be generating traffic at 2 packets per second (period of
> 0.5s).
> 
> And the mobility should be simulated by generating scenario scripts. Do I
> use setdest for this?
> 
> Could anyone kindly give me a simple template or guidelines to where and how
> to set up those parameters in a simulation script file.
> 
> Many many thanks in advance,
> 
> regards,
> vitz

Reply via email to