Master Thesis - 2007

Recently, all optical WDM (Wavelength Division Multiplex) networks are noticed as technology realizing high-speed and large capacity networks. Especially, OPS (Optical Packet Switching) network is considered as promising candidate for next generation network architecture because it has high affinity for existence IP networks. While, optical RAM hasn’t realized yet, so collision avoidance by simple buffering like existent IP networks is difficult. Thereof, high packet loss probability is a serious issue in feasible OPS networks. Now, there are some ways to solve this issue, such as Priority-based Wavelength Assignment (PWA). PWA is realized local load balancing by quota of route and wavelength. These ways, however, do not consider global load balancing, so there is the room for improving efficiency of networks.

Under the background above mentioned, this dissertation proposes new method for global load balancing at practicable processing time. In proposal method, each node observe its traffic load, and setting server calculate all routes and wavelengths all together based on those observed result. While, to calculate strict answer is difficult because this calculation is NP-hard. Thus, in order to calculate by practical time, I adopt an approximation calculated by Genetic Algorithm (GA). Moreover by improvement procedure of GA, raising a probability of escape from local answer and restraining a probability of changing route and wavelength is realized. Simultaneously, I also mention a way of calculating evaluation value of GA suited for proposal method. Finally, proposal method is evaluated by simulations. As result, the effectiveness of load balancing by proposal method is confirmed.