This paper is devoted to dimensioning and optimizing optical buffers for asynchronous, variable length packets in GMPLS-based WDM optical packet-switched networks, which aims to lower network congestion and increase utilization efficiency of optical buffers. In GMPLS-based WDM optical packet-switched networks, the critical issue is how to improve network flexibility and limit network congestion. Comparing to optical buffers with a single operating wavelength, WDM optical buffers in an optical packet switch (OPS) have demonstrated superior performance in solving congestion, which perform buffering by exploiting both time and wavelength dimensions. However, the realistic buffering capacity of optical fiber-delay-line (FDLs) buffers is currently limited. Hence, the algorithm for wavelength sharing and design of the optimal value of the basic time unit of the FDLs in WDM optical buffers are indispensable to shorten FDLs and put OPS into practice. In this paper, the algorithm for wavelength sharing among multiple connections is proposed, which aims to minimize congestion and shorten the queue length. And, in this scenario, the optimal value of the basic time unit of the FDLs is designed. Finally, the proposed algorithm is compared with others in terms of performance.
Optical crossconnects (OXCs) are critical core for provisioning and restoration in mesh wavelength-division-multiplexing (WDM) networks. An increasingly urgent need for large-port-count OXCs severely challenges the current existing OXC technologies. To reduce the crosspoint complexity, we propose an architecture based on 2×2 switching fabrics by integrating the general symmetric (GS) architecture with Clos and Benes switching architectures together. Rearrangeably and strictly nonblocking structures are examined as well as the control algorithm of the rearrangeably nonblocking structure is studied. Then, we present two basic switching fabrics of the simples 2×2 bidirectional OXC utilizing 2D optical MEMS, one of which is used as the basic building block in our proposed architecture is studied. The resulted switch requires (N/2)×[log2(N/2)]×(log2N-1/2) micromirrors, while the switch based on GS architecture needs 2(N/2)2 micromirrors. It is very clear that our proposed architecture reduces the number of micromirrors greatly, especially when N is large. Moreover, theoretical analyses have shown that the resulted switch has the same insertion loss, lower power consumption, and better performance of port-to-port repeatability, comparing to the conventional crossbar switch.