Trace-driven Path Emulation of Satellite Networks using Hypatia

TL;DR

This work extends the Hypatia satellite-network simulator with a trace-driven emulation workflow that records end-to-end path characteristics during non-real-time simulations and replays them in real-time on real systems via the TheaterQ kernel module. By generating Trace Files that capture delay, bandwidth, and queue dynamics, the approach enables running real software and human-in-the-loop tests while preserving fidelity to the simulated network. Three scenarios across different constellations demonstrate high correlation between emulation and full simulation for TCP goodput and RTT, validating the method's effectiveness and revealing practical limitations such as synchronization and lack of simulation feedback. The results suggest trace-driven emulation as a scalable and accurate alternative for protocol development and evaluation in dynamic satellite networks, with future work targeting richer traffic models, feedback loops, and standardization for broader interoperability.

Abstract

The increasing prevalence LEO satellite mega-constellations for global Internet coverage requires new approaches to evaluate the behavior of existing Internet protocols and applications. Traditional discrete event simulators like Hypatia allow for modeling these environments but fall short in evaluating real applications. This paper builds upon our previous work, in which we proposed a system design for trace-driven emulation of such satellite networks, bridging the gab between simulations and real-time testbeds. By extending the Hypatia framework, we record network path characteristics, e.g., delay and bandwidth, between two endpoints in the network during non-real-time simulations. Path characteristics are exported to Trace Files, which are replayed in real-time emulation environments on real systems, enabling evaluations with real software and human interaction. An advantage of our approach is its easy adaptability to existing simulation models. Our extensive evaluation involves multiple scenarios with different satellite constellations, illustrating the approach's accuracy in reproducing the behavior of satellite networks. Between full simulation, which serves as a baseline for our evaluation, and emulation runs, we observe high correlation metrics of up to 0.96, validating the approach's effectiveness. Challenges such as the lack of emulation-to-simulation feedback and synchronization issues are discussed.

Figures (13)

• Figure 1: Visualization of the isl load due to Background Traffic in a Starlink constellation (green = low isl load, red = high isl load) and the network path between a gs in Boston and Paris (blue) at t=100s.
• Figure 2: Visualization of the Background Traffic flows and their total bandwidth during the 200-second simulation.
• Figure 3: Overview of the workflows used in this paper. The left side shows the workflow of a full simulation, the unmodified Hypatia workflow. The right side shows the emulation workflow with the additional real-time step added.
• Figure 4: The emulation setup is conceptually similar to the one presented in ottens2025-2. It consists of three vm managed by a testbed controller. Traffic is routed through an emulation system, where a lkm replays the link characteristics from the Trace Files.
• Figure 5: Scenario 1: Selected properties from the Trace File of the path between Boston and Paris using the Starlink constellation. Orange dots indicate temporary packet losses that occurred during gsl handovers in this scenario. The area on the right shows the time between 105 and 115 seconds in detail, where a congested isl caused packets to accumulate in the queue before the link.