Multiple simulation runs with different parameters

[Daniel Brettschneider]

Hi,

is there a way to automatically spawn a lot of simulation runs with different parameters? Currently I have a config header with some defines to set up my simulation. In OMNeT++ it is possible to create a config file and state something like this.is.some.parameter = {1,2,4..8} and OMNeT++ automatically creates all simulation runs in parallel. Is there a way to achieve this in ns-3 or do I have to do this by hand using command line arguments or the config store?

Regards

Daniel

[Konstantinos]

Dear Daniel,

In order to run independently random simulations, you have to use different RngRun number as explained here 

http://www.nsnam.org/docs/manual/html/random-variables.html#seeding-and-independent-replications

This can be easily implemented with an "external" bash script where you call your simulation, i.e. ./waf --run ..., in a loop and pass the RgnRun in command line  

However, this can be also implemented within your scenario somehow. It will be a bit more difficult though.

[Daniel Brettschneider]

Hi,

I am aware of that but my parameters should not be random. For example I want to define the number of nodes from 1 to 10. Is there a way that ns-3 reads this configuration, starts the 10 simulations with the according parameter, so that I can access it and create the according network topology?

Regards

Daniel

[Konstantinos]

Do you want to have the number of nodes as random variable or fixed?

You can have your parameters fixed, and the randomness with come in places like mobility, start/stop times, backoffs etc.

Also with the bash script, you can have 2 loops, one for changing the RngRun and one to change the number of nodes.

You can use ConfigStore to pass parameters from a file.

http://www.nsnam.org/doxygen/classns3_1_1_config_store.html

[Daniel Brettschneider]

Hi,

I did not find a bash script that does this job so I hacked together one myself. Maybe someone can use this.

It is a python script that sets up some parameters and then starts a lot of simulations in parallel. The first two parameters have to be adapted. I used them to define the number of nodes in my simulation. The third parameter defines the number of repetitions for each value (RngRun creates randomness). The last parameter defines the maximum number of processes that should run in parallel.

The parameters are passed as command line parameters to ns-3. A tutorial on how to use them is available here: http://www.nsnam.org/docs/release/3.9/tutorial/tutorial_22.html

#!/usr/bin/python

import sys

import os

from multiprocessing import Pool

def start_simulation(data):

run = data[0]

sub_run = data[1]

homes = data[2]

os.system('./waf --command-template="%%s --run_number=%d --sub_run_number=%d --grid_homes=%d --RngRun=%d" --run citygrid' % (run, sub_run, homes, sub_run))

if len(sys.argv) != 5:

print "usage: ./parallel [max_homes] [step] [sub_runs] [processes]"

sys.exit(0)

max_homes = int(sys.argv[1])

step = int(sys.argv[2])

sub_runs = int(sys.argv[3])

processes = int(sys.argv[4])

# create params

params = []

run = 1

for i in range(0, max_homes+step, step):

homes = i

if homes == 0 and step > 1:

homes = 1

elif homes == 0 and step == 1:

continue

for j in range(sub_runs):

params.append([run, j+1, homes])

run += 1

# run

pool = Pool(processes=processes)

pool.map(start_simulation, params)

Regards

Daniel 

[Mahmod Ahmad]

 is it solved to simulate nodes number from 1 to 10? please answer