The Knowledge Plane (KP) architecture enables next-generation computer networks to be self-describing, self-analyzing, and ultimately more self-managing. This one-year project is laying out an initial high-level design for the KP and applying it to focused test problems, with the aim of identifying key steps towards full development and deployment of a KP-enabled network. A fully KP-enabled network will be able to detect and diagnose its own failures and faults, and often advise on its own repair. The project would first define a three-part KP architecture consisting of (1) an agent system to decompose distributed and partitionable learning and analysis, (2) a knowledge representation framework and (3) knowledge routing and management. Key design criteria are scaling, distribution, and reconciliation of incomplete or incorrect knowledge. The KP postulates that sharing knowledge across multiple problem domains will strengthen solution performance in each domain. To motivate the architecture design and validate this postulate, the research focuses initially on a model-based hard fault diagnosis application, which is then extended to discovery of security failures. Following preliminary fault diagnosis work at MIT, the PIs focus on ISI's DETER network security testbed as a test environment for diagnosing hard fault and security-related configuration failures. DETER offers a controllable and constrained to simplify initial research, while being sufficiently large and complex to motivate many of the KP mechanisms required for a commercial Internet. The intended outputs of this project are (1) an architecture, including principles, goals, and an initial design, (2) a specific application of these to the separate yet overlapping problems of hard fault and security failures, and (3) identification of key questions to be addressed in the next steps of the research. Results will be disseminated through incorporation in the operational DETER testbed, publication of papers and related code, and PI participation in NETS-FIND community.
