RE: NS versus OPNET Modeler

[Xin Wang <xwang@ctr.columbia.edu>]

On Mon, 17 Jan 2000, James Scott wrote:

> hi heather,
> 
> > I am negotiating a purchase of the OPNET Modeler software 
> > and a friend reccommended NS.  Any old or current OPNET 
> > users? Who could give me feedback?
> 
> i have used both opnet and ns.  the question of which is
> the best tool would depend on the problem you're trying
> to solve and the skills of the person who is doing the
> work.
> 
> the process model OpNET provides is a powerful tool,
> providing sort of a 4GL for building simulation models
> of specific nodal processes.  additionally, the opnet i/o
> is top notch.  and, if your work involves modelling ATM
> nodes, opnet is clearly the way to go.  also, mil3 has 
> recently provided a VHDL i/f for OpNET, which would be 
> quite handy for cosimulation applications.
> 
> i've found NS to be superior in the area of source traffic
> generation.  while OpNET does provide a flexible mechanism 
> for generating various types of source traffic, NS has a 
> larger number of source traffic generators, in addition to 
> contributed models for higher layer protocols (HTTP, etc).
> however, since OpNET provides an open API for source traffic
> generation, there is nothing prohibiting you from using NS
> generators to stimulate an OpNET model, depending on the
> level of sophistication possessed by the developer doing
> the work.  in the past, i've written pareto models to
> stimulate OpNET process models.  had i known of NS at the
> time, i could have saved some time by leveraging their
> pareto model <shrug>.
> 
> in the area of routing, NS provides static and session-based 
> routing using dijkstra, as well as dynamic routing using a 
> distributed bellman-ford.  by comparison, the OpNET tools 
> provides just static routing.


Opnet also provides RIP and  OSPF routing, as well multicast routing 
recently. It is a more powerful simulation tool, the code is 
well organized and the tool is good for demo. 
But it has a bigger learning curve as trade-off.
 
> so, in summary, if you are modelling an IP network
> and interesting in things like queueing behavior and
> the interaction of higher-layer protocols on network
> performance, go with NS.  you can use NS to model
> specific behaviors of proprietary h/w, but its perhaps
> not for the faint-of-heart.
> 
> there are probably other areas where NS would be preferred,
> such as characterization of IP multicast and/or wireless IP,
> but i haven't had experience with these areas so will reserve
> comment.  the above is just one man's opinion.  both tools
> are QUITE useful, depending on the task at hand (as well as
> the size of your budget, and skills of your development crew).
> 
> good luck!
> 
> cheers,
> 
> james
>