2.2 Use cases

To introduce the design of ns-3, we first review design issues and usage models that have arisen with ns-2, and mention trends in simulation use within the networking research community.

• Model extensibility. Most research users want to extend the simulator by writing new simulation scripts, modifying existing models, or writing new models. To facilitate model modification, ns-3 continues the use of object-oriented design with polymorphic classes, allowing users to subclass the aspects that they wish to modify. To facilitate the addition of new models, ns-3 adopts a component-based architecture for compile-time or run-time addition of new models, interface aggregation, and encapsulation, without requiring modification of the base models of ns-3.
• Simulation code reuse. Many users start their work with ns-2 by adapting existing code. Some common code is written in terms of base-class object pointers, allowing for run-time substitution of subclassed objects. ns-3 will use several techniques to facilitate simulation code reuse, such as inheritance to extend existing classes, the provision of (extensible) stock topology objects, simulation frameworks that are easily modifiable, an example script repository, and a system for run-time configuration of classes and default values.
• Run-time configuration. ns-3 provides a flexible technique to allow users to redefine default values and class types without recompiling the simulator. The default value database is integrated with a command-line argument parsing facility, making all the variables configurable from the command-line as well.
• Tracing. ns-3 features a callback-based approach to tracing that decouples tracing sources from tracing sinks and that is focused on flexibility for the user. Packet traces will be made available in libpcap format, to allow for post-processing tools built around that trace format. Built-in statistics will also be widely available.
• Scaling. ns-3 will include techniques for improving the scalability of simulations, including distributed simulation techniques introduced with PDNS and GTNetS, scalability techniques introduced for wireless simulations such as caching of computationally-intensive results, and flexibility in tracing infrastructure (to avoid large traces).
• Software integration. ns-3 is oriented towards the reuse of existing software such as routing daemons, applications, and kernel code. The design is built around encapsulation techniques that decouple the interface from the implementation, an architecture that better mirrors how real-world devices are built (e.g., explicitly handling multiple interfaces per node), and an abstraction library that allows implementation code to run in both real and simulated environments.
• Network emulation. Increasingly, network research that involves simulation also includes an experimental component, with facilities such as PlanetLab, Emulab, and ORBIT. Researchers would like to more easily move between simulation and experimental domains. The ns-3 design is intended to facilitate this interaction between simulation and experiments, with a packet design oriented towards serialization and deserialization, and encapsulation techniques that will allow real application and kernel code to run in the simulator, thereby improving traceability to real-world implementations.
• Scripting The primary ns-3 user interface at present is a C++ ``main'' program, and we expect that C++ will continue to be a preferred language for many users. However, ns-3 will also feature Python bindings allowing for users to define scripts and replaceable components in Python.

We organize the rest of the discussion in this chapter as follows:

1. Class and object design
2. Memory management
3. Configuration
4. Tracing
5. Scaling
6. Emulation
7. Scripting

The next chapter goes into more detail on the Node, Channel, and Packet object designs.