While network simulation is widely used for evaluating algorithms such as congestion control, configuring a simulator to faithfully mimic the target network is challenging. In the iBox, or Internet in a Box, project, we seek to address this challenge with a data-driven approach. We focus on the goal of “learning the network” to recreate real-world, end-to-end network path behaviour, based on input-output packet trace telemetry. Using a combination of a simplified network path model and machine learning, we seek to recreate the end-to-end behaviour of a path in the target network when presented with a new input. Our ultimate goal is to enable the realistic testing and evaluation of network protocols and applications in a simulated environment.
Please visit the iBox Github repo (opens in new tab) for data-informed network simulation profiles derived using iBox.