Packet-level network simulation provides high-fidelity modeling but suffers from severe scalability bottlenecks. Existing scaling approaches remain inefficient for modern data-center and AI-training networks. Spatial parallelism requires substantial hardware resources, while temporal-skipping approaches become less effective under bursty traffic. We observe that homogeneous data-center deployments introduce substantial execution redundancy during simulation.

This paper presents EmuFork, a novel emulation paradigm that eliminates redundant initialization by mimicking OS-level fork semantics. EmuFork captures the post-initialization state of a single template router and rapidly restores all subsequent instances from this shared snapshot via Copy-on-Write. By bypassing redundant computations and resolving cloning conflicts, EmuFork accelerates large-scale emulation by up to 86% and reduces peak memory usage by up to 30%.

This paper introduces Galvatron, a domain-specific language designed to simplify transport layer programming. Galvatron obscures the underlying data structures and operations, and exposes parametric processing stages to the developer, thus significantly reducing coding effort. Our evaluation demonstrate Galvatron’s ability to encapsulate complex semantics of 4 real-world transport protocols using merely 2% of the code compared to native implementations.