Routing Control of Packet Flow by using Statistical-Physical Methods

点击翻译按钮获取中文摘要

1. Study on packet routing controlWe have proposed statistical-physical models which are decentralized, autonomous and adaptive for packet flow on a large computer network. An energy function is defined in order to express competition among queue length, distance to destination of a packet. The packet routing control is performed by using the energy function. We have proposed several formulations using neural networks, Ising models, or reinforcement learning systems. It has been shown that these models give good performance for packet routing control.2. Study on designing of optimal network strictures for packet flowWe have investigated an optimal network structure for packet flow on large networks by defining a cost function which represents efficiency of the packet flow. As a result, we have found that random networks are optimal in the case where a destination is determined randomly, and networks with scale-free property are optimal in the case where the buffer-size distribution has a power-law form and a destination is determined with the probability which is proportional to the buffer size of the node.3. Study on generating mechanism of scale-free networksWe have studied on possibility of self-organization of nongrowing networks into scale-free networks. By introducing fitness parameters in rewiring processes, we have shown that networks with scale-free property are generated also in the nongrowing case. The same behavior has been observed in the nongrowing model for bipartite graphs.