BrowsEm: Model-based Web Site Loading Emulation

Continuous research and engineering efforts aim to improve the performance of web networking protocols, such as TCP, QUIC, TLS, and HTTP. Performance measurements are conducted to assess the impact of changes to these protocols, their implementations, or the underlying network infrastructure. However, this is a challenging task due to the increasing complexity of web site deployments and browser and server software. In this work, we introduce BrowsEm, a web site loading emulator capable of reproducing page loading workloads. Its underlying model is based on data scraped from real web sites and takes network path characteristics, protocols, dependencies between HTTP transactions, and their individual timing aspects into account. Modularly structured and using wide-spread libraries like libcurl, it allows for testing a wide range of emulation parameters such as network conditions or protocol implementations and obtaining reproducible results. Our evaluation shows that the emulation introduces a relative error smaller than ± 0.25 for 80% of observed page load times. In a measurement campaign, we find that the model shows suitable robustness for artificially changed parameters.