Networking and Cybersecurity


Scientific Experiment Modeling

Valid scientific experiments are required to accurately evaluate and assess networking systems. Conducting these experiments necessitates modeling multiple complex network, environmental, traffic, and behavioral effects and systems. Unfortunately, the majority of experimenters are not experts in all the domains required to conduct these types of experiments. For example, an expert in loss-resilient transport protocols may not be an expert in generating accurate loss and jitter models for evaluation. Technologies must be made available that can help capture the knowledge of one dolman expert and enable another to utilize it without becoming an expert themselves.

The Networking and Cybersecurity Division has undertaken a significant research effort to address this issue. Part of this effort is the design of a system which enables modeling experts to contribute tools that are easily usable by other experimenters. Experimenters search through the repository to find modeling capabilities for their specific needs. Additionally, other experts can use these tools to generate models of their specific corner of a field and contribute those models back to the ecosystem. For example, an engineer at CenturyLink could generate a model for their network using the available modeling tools. This enables an experimenter to evaluate their system on an accurate model of the CenturyLink network without requiring detailed knowledge of that network or the skills required to model it. 

Capturing and disseminating this knowledge is only part of the overall problem to be addressed. Experimenters must also be educated on what models are actually useful for their experiments. Experimenters typically use overly simplified models or desire perfect models—the former leading to invalid results and the latter being superfluous to the goals of the experiment. To this end, we are working with experimenters to help understand their experimental goals and constraints, and select appropriate models based on them. Our ultimate goal is to encode this information directly within the experiment description and have the experimental infrastructure engine suggest potential models that may satisfy the experimenters goals and constraints. Multiple research efforts within the Networking and Cybersecurity Division are exploring solutions to this problem in parallel.