| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 96 === The demand for higher communication bandwidth has been growing at a rapid pace led by diverse network services. It is necessary to improve the node throughput to achieve high-speed data transmission. Wavelength Division Multiplexing (WDM) is a promising approach that can provide an aggregate throughput of the order of terabits per seconds. It is widely favored as a practical solution to fulfill the growing communication demand. For enhancing the transmission rate from electronic domain, the future WDM systems are likely to be implemented by all-optical components.
Packet contention resolution is a major issue in optical packet switching networks, especially in asynchronous switching networks. In asynchronous switches, packets arrive in continuous-time fashion, and there may be more than one packet destined to the same output port simultaneously. In the electronic domain, this problem is typically solved by buffering the contending packets. However, RAM-like buffering is not yet available in the optical domain. In optical networks, fiber-delay-lines (FDLs) are used to delay packets for a fixed amount of time. Different from Wavelength Conversion technology, fiber delay line provides flexible buffering strategies, especially when the number of wavelengths is limited.
A limitation of optical buffering is that FDLs may be bulky, since the length of a fiber is directly proportional to the travailing time of light in the fiber. Efficient buffer architecture and management play an important role in reducing packet blocking probability and the number of buffers. In the thesis, we investigate three buffer management schemes for feedback-type optical packet switches: they are Available Unused Delay-Line scheme (AUDL), Available UnScheduled Delay-Line scheme (AUSDL) and Available UnScheduled Delay-Line with Group selection (AUSDL-G). The simulation results show that AUDL scheme yields the highest utilization rate of FDL. However, the complexity for implementing the AUDL scheme is considerably high and may become infeasible in practice. To reduce the complexity of the AUDL scheme, we further propose AUSDL and AUSDL-G schemes. Based on the simulation results, we find that the AUSDL and AUSDL-G schemes perform as well as the AUDL scheme in terms of packet blocking probability, but with much lower computation complexity.
|
|---|
