Programmable network devices were originally conceived as platforms for implementing new packet-processing functions. But what if we programmed them not only to build the network under study, but also to build the testbed around it?
This keynote explores “Software-Defined Inception”: the use of programmable switches and SmartNICs as building blocks for realistic, reproducible network experimentation. It presents an evolving open-source toolbox that combines hardware-accelerated network emulation, traffic generation, and protocol testing. Tools such as the P4 Programmable Patch Panel (P7), PIPO-TG, and P4R demonstrate how programmable hardware can reproduce configurable topologies and link conditions, replay packet traces, and generate challenging traffic patterns—including stateful flows—at line rate.
Drawing on experience with these tools in research and education, the talk discusses both the opportunities and practical challenges of turning data-plane devices into experimental infrastructure. These include preserving timing fidelity, sharing limited hardware resources, composing experimental functions with the system under test, and making experiments portable and reproducible across platforms.
Finally, the talk offers a first look at SPINE, ongoing work toward a practical network tester spanning programmable switches and SmartNICs. SPINE builds on the lessons of the existing toolbox to explore how limited physical infrastructure can represent richer experimental environments while retaining hardware-level fidelity.
The broader argument is that the next frontier for programmable data planes extends beyond implementing individual network functions. By programming the experimental environment itself, we can create compact, flexible, and high-performance testbeds in which future network protocols and systems can be built, stressed, observed, and validated under realistic conditions.