EpiNet: a simulation framework to study the spread of malware in wireless networks

Karthik Channakeshava, Deepti Chafekar, Keith R. Bisset, V. S. Anil Kumar, Madhav V. MaratheSimuTools 2009 · OpenReview

Abstract

EpiNet is a high-performance simulation framework designed to study how malware spreads across realistic, city-scale wireless networks. By replacing oversimplified random movement models with activity-based human mobility data, it achieves orders-of-magnitude speedups over traditional simulators while providing far more accurate predictions of epidemic dynamics.

Introduction

As smartphones and wearable devices become ubiquitous, they create a massive, ever-shifting "contact network" ripe for malware like Bluetooth worms. To defend against these threats, researchers need to predict how an outbreak might evolve. However, existing tools face a "Goldilocks" problem: analytical models are fast but oversimplified, while packet-level simulators (like NS-2) are highly detailed but incredibly slow, taking days to simulate just a few hundred devices. Furthermore, most studies rely on unrealistic "Random Waypoint" (RWP) models, where nodes move like billiard balls. In reality, human movement is driven by schedules—going to work, shopping, or school—which creates fundamentally different network structures and infection patterns.

Methods

The authors developed EpiNet, a parallel distributed discrete event simulator that bridges the gap between speed and realism through several novel approaches:

Activity-Based Mobility

Instead of random movement, EpiNet uses the Simdemics framework to generate synthetic urban populations based on actual census data and time-use surveys. This creates realistic "visits" to locations with specific arrival and departure times. This temporal flow is critical for capturing the dynamic nature of contact networks in a city, as the density of potential hosts fluctuates throughout the day.

Temporal distribution of arrivals and departures in the synthetic urban population.

Sublocation Modeling

To handle the computational load of proximity-based connections, the authors divide locations into grids. This "coarse-grained" approach allows the simulator to quickly identify which devices are near each other without checking every possible pair, significantly reducing the complexity of neighbor discovery in dense environments.

Sublocation modeling grid for proximity-based connection handling.

Probabilistic Timed Transition System (PTTS)

Rather than simulating every individual radio packet, EpiNet abstracts Bluetooth worm behavior into a state machine. This model captures the timing of inquiry, paging, and infection phases based on statistical distributions derived from detailed low-level simulations, allowing for rapid execution without sacrificing fidelity.

Key Insights and Results

EpiNet demonstrates significant advantages in both performance and accuracy over traditional modeling approaches:

  • Massive Speedup: In a head-to-head comparison with 500 devices, EpiNet completed in 14 minutes what took NS-2 two full days to simulate.
  • High Fidelity: Despite the abstractions, EpiNet’s infection curves closely matched the results of detailed packet-level simulations, proving that bit-level simulation is not always necessary for accurate epidemic modeling.
  • Scalability: The authors successfully ran simulations involving 20,000 devices across an 8-hour window in a downtown Chicago environment, a scale previously impossible for detailed network simulators.
  • Timing and Density Sensitivity: The "seed time" (when the first device is infected) drastically changes the outcome. Infections starting during peak commute hours (e.g., 8 AM) spread much more effectively than those starting later. Additionally, doubling device density led to a 5x increase in the rate of new infections during the first hour.

Conclusion and Future Work

EpiNet provides a robust and scalable framework for studying urban-scale epidemics. By incorporating realistic mobility and efficient sublocation modeling, it offers a more accurate tool for testing defense mechanisms. Future work will focus on incorporating device heterogeneity (different OS versions and user settings), integrating non-proximity spread (SMS/MMS), and testing intervention strategies such as early detection systems or digital vaccinations. The ultimate goal is to scale the simulation to represent entire metropolitan areas with millions of individuals.

Summary updated

BibTeX
@inproceedings{channakeshava2009epinet,
  author = {Karthik Channakeshava and Deepti Chafekar and Keith R. Bisset and V. S. Anil Kumar and Madhav V. Marathe},
  title = {{EpiNet: a simulation framework to study the spread of malware in wireless networks}},
  booktitle = {SimuTools},
  year = {2009},
  url = {https://openreview.net/forum?id=IvOACemAfh},
}
Figure