Abstract
Opportunistic networks (OppNets) rely on intermittent connectivity to transfer data, traditionally modeled at the packet level. However, as data volumes increase, discrete-event simulations become computationally expensive. This paper presents a hybrid modeling approach using stochastic HYPE, a process algebra that treats data as a continuous fluid governed by Ordinary Differential Equations (ODEs). By combining fluid flows with discrete stochastic events, the framework enables efficient performance analysis of complex scenarios, such as mobile data ferries in wildlife monitoring. Results demonstrate that this fluid abstraction maintains high accuracy while significantly reducing simulation time compared to traditional methods.
Introduction
Opportunistic networks (OppNets) are characterized by their lack of persistent end-to-end connectivity. In these environments, nodes—such as sensors in remote areas or mobile devices in urban settings—must utilize a "store-carry-forward" paradigm. Data is stored locally and transferred only when a communication opportunity arises.
Traditionally, performance modeling for OppNets has relied on packet-level simulations. While precise, these methods face a scalability crisis as data volumes grow. For instance, modeling a network of high-definition video sensors generating gigabytes of data requires tracking millions of individual packets, a process that is both computationally intensive and slow. To address this, we propose a hybrid approach that abstracts data as a continuous fluid, allowing for rapid performance evaluation without sacrificing the nuances of intermittent connectivity.
The Stochastic HYPE Framework
The core of this approach is stochastic HYPE, an extension of the HYPE process algebra designed for hybrid systems. Unlike traditional simulators that count discrete packets, HYPE models data as a continuous quantity whose rate of change is defined by Ordinary Differential Equations (ODEs).
Fluid Abstraction and Compositional Modeling
In HYPE, the system is modeled through "flows" or influences. Instead of manually writing complex global equations for a network, a modeler defines local behaviors. For example, a sensor's data accumulation and its subsequent upload to a mobile ferry are defined as separate influences. When these components are composed, the global behavior of the system—such as buffer occupancy over time—emerges automatically from the underlying semantics of the language.
Handling Discrete and Stochastic Events
While data flow is continuous, the network topology is discrete. HYPE manages this through two types of events:
- Urgent Events: These occur when a system variable hits a specific threshold, such as a data buffer reaching maximum capacity.
- Stochastic Events: These represent random occurrences, such as a mobile ferry coming within range of a sensor. These events are modeled using probability distributions (e.g., exponential distributions for contact rates).
Case Study: Wildlife Monitoring
To evaluate the efficacy of the HYPE framework, we developed a synthetic case study based on a wildlife monitoring scenario, similar to the ZebraNet project. The setup included ten stationary video sensors, each with a 250MB storage capacity. These sensors record video three times daily, and a mobile "ferry" is tasked with collecting this data and delivering it to a base station over an eight-hour window.
We tested several variables, including ferry buffer sizes (ranging from 600MB to 1400MB) and different mobility models:
- Fixed Routes: The ferry follows a predetermined, cyclical path.
- Random Routes: The ferry moves between sensors based on stochastic meeting rates.
Key Results and Insights
Simulation Efficiency
One of the most striking advantages of the fluid approach is its speed. A complex simulation of an eight-hour scenario involving ten sensors and a mobile ferry was completed in just 6.5 seconds on a standard laptop. This efficiency is achieved because the simulator integrates ODEs rather than processing millions of discrete packet-transfer events.
Routing and Fairness
The analysis revealed that routing strategy significantly impacts network "fairness." Fixed routes (cyclical paths) consistently outperformed random routes. In random mobility models, certain sensors were often bypassed for long periods, leading to buffer overflows and massive data loss, even when the average visit frequency remained high.
Buffer Management Policies
The study also compared different ferry behaviors. For lower buffer capacities, a "return to base when full" policy was essential. Without this logic, ferries would continue to visit sensors despite being unable to collect new data, leading to wasted energy and increased data drops at the sensor level.
Conclusion and Future Work
Hybrid performance modeling via stochastic HYPE provides a powerful alternative to traditional packet-level simulation for high-volume opportunistic networks. By treating data as a fluid, researchers can gain deep insights into system bottlenecks and routing efficiency in a fraction of the time.
Future developments will focus on moving from abstract meeting rates to 2D spatial models, allowing HYPE to explicitly model continuous movement within a coordinate system. Additionally, we aim to modularize the simulation software, creating parametric templates that allow users to easily deploy and test various node configurations and mobility patterns in larger, more complex network environments.
Summary updated