Networked systems researchers face challenges in their development process, transitioning from local testing to scaling experiments on public testbeds. The manual creation of ad-hoc scripts for parameter exploration and analysis leads to inefficiencies, hindering the development process and reproducibility. In this poster, we present a Network Performance Framework(NPF) that orchestrates experiments automatically, from the step where a prototype is ready for its first run to the final graphs one can embed in a paper. The framework generates graphs for each metric, interactive web pages and Jupyter notebooks that allow users to interact easily with the experiment’s results. NPF is easy to start with, as the simplest test description file is just a list of bash commands to run. Gradually, NPF makes it easy to deploy an experiment over local and remote testbeds, enabling complex deployment, synchronization, and data collection. Each experiment can grow a set of factors ( e.g., number of threads, buffer size, packet length, packet rate, . . . ) that NPF uses to orchestrate the experiment and collect performance metrics ( e.g., throughput, latency, ...) of different configurations. NPF helps the researche in building their experimental design to select meaningful factors, automatically finding regions of interest and easily cutting through multi-dimensional data.
Barbette, T. (2024). Poster: NPF: orchestrate and reproduce network experiments. 2024 ACM Conference on Reproducibility and Replicability, Rennes, France. https://hdl.handle.net/2078.5/271377