This project focuses on the following challenges in operating network systems:

• In network design, build, and operations, you must consider how multiple devices interoperate. With growing scale and the introduction of virtualization, it is becoming harder to predict “what will happen.”

• Pre-production testing is required to clarify “what will happen,” but fully reproducing a network environment made up of many devices is difficult. In practice, testing must be limited in size and functionality and requires cost (equipment and time). As a result, it remains difficult to eliminate problems that are “only discovered in production.”

 

To address these challenges, we take the following approach:

• We model, in a machine-readable form, the information operators use to understand and operate networks—configuration, structure, and topology. By making “configuration information” machine-processable, we expand and improve the scope and precision of what can be engineered.

 

Building on this idea, we have conducted experiments and explored use cases such as:

• Modeling network “configuration information” so that static checks (e.g., inter-device configuration conformance) can prevent basic misconfigurations.

• Using a simulator derived from the “configuration information” to simulate portions of the production network in advance and prevent incidents.

• Recognizing that simulation has limits, using container-based networking products (CNFs) to reproduce an equivalent test environment from the same “configuration information,” enabling broader functional testing and preventing incidents.

• Performing validation inside the reproduced network and automatically generating (a) diffs between the updated and original “configuration information,” and (b) the concrete configuration changes that should be applied back to the production network.

• Using CNFs—which are lighter-weight than traditional testbeds (hardware- or VM-based)—to build larger (more nodes) environments more quickly and accelerate the verify-and-feedback cycle.

 

A demo video is also available.

Reference (slides): “Would you be happy if you could *copy* your entire production network into a virtual environment?” / JANOG 51 (Speaker Deck)