[Ns-developers] ns-3 and future development

[Sam Jansen]

On Tue, 2006-01-17 at 20:05 +0100, mathieu lacage wrote:
> hi daniel,
> 
> Daniel Mahrenholz wrote:
> [...snip...]
> > There is one thing that would be very useful (at least for me) - 
> > parallelization/distribution of independent replications and 
> > aggregation of results/traces in one place.
> 
> 
> I see. Yes, it would be useful to you and other users (me included). 
> However, I see this sort of problem as being decorelated from the 
> simulator itself: it seems to me to be a matter of writing the right 
> scripts to control the simulator. I wonder if such scripts could be made 
> generic enough to be useful to more than one user.

This is an interesting problem. I've probably had about 3 different
iterations of scripts and programs that control ns-2 over the course of
me using it. I've talked to a few others who have also written their own
scripts to do the same sort of replication tasks when using ns-2.

It would be nice to have some standard addition to ns-2 which handled
this. My current solution is much more engineered than my previous
hacks, and has been used on up to 100 machines at once to distribute a
large set of simulations that took several weeks to run. It
automatically sends the most current version of simulation files out to
simulation machines and returns the results back to the central server
machine. An ugly yet functional web interface allows control, as well as
command-line programs.

It also makes some assumption about the setup of simulations and such,
meaning it is perhaps not as general as it could be. I'm not sure how
much more general it could be made, though. 

I've vaguely wondered if my work was useful to anyone else during the
development of it, but never really got around to announcing it, I just
assume everybody else has written their own version of the same thing. 

If anybody lurking on this list thinks it sounds useful to them, give me
an email and I'm happy to tar up the code. Otherwise I think a nice
standard way of distributing independent replications is something the
next version of ns should definitely solve.

Cheers,
Sam

-- 
Sam Jansen                                               sam at wand.net.nz
WAND Network Research Group                           Waikato University