Robust Transfiguring Network Protocols.
Final technical rept. Mar 92-Mar 95,
SRI INTERNATIONAL MENLO PARK CA
Pagination or Media Count:
In RTNP, we have developed a protocol that uses two artificial intelligence methods, neural networks and evidential reasoning, to recognize and predict adverse network conditions, and that uses fuzzy logic to dynamically control the parameters of a tunable routing protocol in response to the perceived environment. Examples of the tunable protocol parameters are 1 a parameter that controls the degree to which traffic is spread over multiple paths 2 a link bias parameter that, when large, increases stability by forcing traffic over minimum-hop paths and 3 a parameter that determines how often routing updates are sent. Examples of measurements used to recognize adverse conditions are 1 congestion 2 probability of a successful transmission on a link 3 jamming characteristics and 4 degree of routing oscillations. Neural network methods were developed for predicting link-states and congestion, based on network measurements and estimates. These methods were shown in simulations to predict link states and queuing delay much more accurately than other methods.
- Computer Systems