Abstract:

The traffic optimization under the inaccurate network information was discussed. Defining the virtual capacity, a model was made on the local state information; a distributed real-time and unsupervised learning algorithm was proposed, which can learn the range and frequency of the network traffic variation and then choose adaptively whether or not and how to adjust the traffic distribution between parallel paths. This algorithm can optimize the network traffic efficiently. It need not to make statistics, refresh and keep all kinds of state information and traffic matrix in the network.

Key words: traffic engineering, local state information, inaccurate, real time, unsupervised learning