6G-PATH [6G-RESCUE]: Container-Based Edge AI Deployment over Cloud and Edge Infrastructures (dataset)

This dataset contains TCP header captures collected during the deployment of a video object recognition application on a Jetson-based edge AI device, as part of the 6G-RESCUE project under 6G-PATH. The captures contain TCP-layer data exchange across two deployment strategies and two network conditions, providing a controlled benchmark for evaluating the impact of network latency and deployment method on application delivery to edge AI devices.

Two deployment strategies are compared: Type 1, in which the device downloads a complete pre-built container image (~135.5 MB) from the server, and Type 2, in which only a lightweight build script (~15 kB) is transferred and the application is compiled on-device. Each strategy is evaluated under two network conditions: a cloud scenario with a 50 ms one-way delay (2 routing hops) and an edge scenario with a 7 ms one-way delay (1 routing hop), resulting in four captures in total.

All captures use a 96-byte snapshot length, preserving the full TCP/IP and Ethernet headers while truncating the payload.

The dataset is intended to support research into TCP performance modelling, edge vs. cloud deployment trade-offs, and network-aware application deployment strategies for edge AI workloads.