Meeting the agreed quality of service in a resource crunched data network is challenging. An intelligent element is required to carry out the activities involved. The inferences drawn with different rules need to be merged. Agents are useful for handling this responsibility in data networks and help in resource sharing. An agent is basically an entity that can be viewed as perceiving its environment through sensors and acting upon its environment through effectors. To handle the network traffic, the agents acquire the traffic status and provide the information on the availability of resources to the source of the traffic. Hence the study on agent communication has become important. Intelligent agents continuously perform the activities including perception of dynamic conditions in the environment, reasoning for interpretation of the perceptions, solve problems, draw inferences and determine actions.
TopBackground
The present day network traffic supporting the multimedia data with different QoS constraints have to be routed across the network in real time. Conventional techniques cannot catch up the fluctuations in the traffic making it necessary to use an intelligent element to memorize the changes. A comparison of different techniques used for providing the QoS is discussed in (Manjunath,R., & Shyam,V., 2008). Although a feedback based controller such as Random early detection (RED) provides the congestion status information (Hollot, C., Misra, V., Towsley, D., & Gong, W, 2001) to the data source to alter the transmission rate subsequently, the technique is less adaptive for the fast changing network traffic. Hence, intelligent elements are required to predict the traffic in advance to provide sufficient time for the sources to act.
Neural networks exhibit massive parallelism making them ideal for real time application s. In any system making use of neural networks, when such an element is transferred to silicon the resources such as Buffer size, speed of decisions, area of the associated circuitry etc put stringent constraints making the algorithms of implementation competitive.
The traffic controller proposed here is based on the agents that implement a broker based model. The algorithm employs neural networks to compute desired transmission rate of the source in order to prevent congestion in the subnet. Obtained results prove that the neuro-computing approach is better than the conventional one. This is possible because of the unique learning and memorizing capabilities of the neural network based on the previous experiences.