This thesis proposes the use of real world network stacks instead of protocol
abstractions in a network simulator, bringing the actual code used in
computer systems inside the simulator and allowing for greater simulation
accuracy. Specifically, a framework called the Network Simulation
Cradle is created that supports the kernel source code from FreeBSD, OpenBSD
and Linux to make the network stacks from these systems available to the
popular network simulator ns-2.
Simulating with these real world network stacks reveals situations where the
result differs significantly from ns-2's TCP models. The simulated
network stacks are able to be directly compared to the same operating system
running on an actual machine, making validation simple. When measuring the
packet traces produced on a test network and in simulation the results are
nearly identical, a level of accuracy previously unavailable using traditional
TCP simulation models. The results of simulations run comparing ns-2 TCP
models and our framework are presented in this dissertation along with
validation studies of our framework showing how closely simulation resembles
real world computers.
Using real world stacks to simulate TCP is a complementary approach to using
the existing TCP models and provides an extra level of validation. This way of
simulating TCP and other protocols provides the network researcher or engineer
new possibilities. One example is using the framework as a protocol
development environment, which allows user-level development of protocols with
a standard set of reproducible tests, the ability to test scenarios which are
costly or impossible to build physically, and being able to trace and debug
the protocol code without affecting results.
