Overview

The Internet today is a visible part of daily life in research , business, home, and the military. It promises to also become an invisible, background fabric that supports interaction between devices and people as never before. As use of the Internet expands , improving reliability is the key challenge that must be met for continued progress. Specifically:

• Understanding protocol and network operation under extreme conditions,
• Prediction, detecting and recovering from implementation bugs, and
• Predicting and responding to rapid changes in traffic patterns

Meeting these challenges requires advances in network measurement, modeling, and simulation technology. We propose to address these challenges with network measurement and simulation incorporating innovations in four areas:

• Analysis-driven failure exploration: We will combine analytic techniques (such as fluid flow modeling and stability theory) and simulation approaches(such as validated abstraction and exploratory modularity) to efficiently explore, predict, and track protocol behavior under and recovering from stress.
• Fast-turn-around models as simulation input: We will develop tools and approaches that integrate multi-point network measurements to rapidly generate compact, accurate, application-level models that are accurate across a wide range of time-scales.
• Application of these models to failure prediction: We will use these models to develop network early-warning systems that will predict imminent overload and other network failure conditions, enabling preemptive corrective action and active failure avoidance.

Successful outcome of this work will provide profound new capabilities to the networking community:

• Fundamental new understanding of protocol limits, operation at those limits , and recovery when those limits have been exceeded.
• Early-warning systems that can detect potential network failure and trigger automatic or manual corrective action.
• Tools that can automatically and rapidly produce network models as required by the above two capabilities
• Methodology and libraries for emulation-based conformance testing of protocol, implementations, and models