Network experimentation tools are crucial for developing and testing distributed systems, but current state-of-the-art simulators face inefficiencies at large scales or architectural hurdles. This paper introduces Phantom, a novel tool designed to overcome these limitations.

Phantom directly executes unmodified applications as Linux processes within a discrete-event network simulation environment. It achieves this by innovatively synthesizing efficient process control, system call interposition, and data transfer methods to seamlessly integrate these processes into the simulation. This approach allows for more realistic and scalable network experiments.

Evaluations show Phantom significantly outperforms existing tools. It is up to 2.2x faster than Shadow, up to 3.4x faster than NS-3, and up to 43x faster than g Ra IL in large P2P benchmarks. In large Tor network simulations, Phantom offers performance comparable to Shadow. These results highlight Phantom's potential to advance the field of network simulation and distributed systems research.