This paper considers multicast routing and wavelength assignment (MCRWA) in a wide area all-optical network. We present static and dynamic MCRWA algorithms, and evaluate their blocking performance under various fanout splitting policies. Through simulations and comparisons with a lower bound, we demonstrate that dynamic MCRWA can perform significantly better than its static counterparts. We then quantify the throughput gain due to link sharing in multicasting. We also present an analytical model for estimating the blocking performance of the static MCRWA algorithms. The performance improvement with multifiber networks and with wavelength conversion are also investigated.

We analyze the shortest path lengths between node pairs of real optical transport networks. From the analysis, we find that Johnson SB distribution is suitable for the shortest path length modeling. The validity of the distributions is evaluated in terms of the Kolmogorov-Smirnov (KS) statistic. Johnson SB distribution provides an average KS statistic of 0.0423, which indicates its good accuracy. We also show that the key parameters of the shortest path lengths, such as the mean, the median, and the standard deviation, can be estimated from the convex area of the network. We develop the proposed Johnson SB distribution model for the shortest path lengths using the basic information of the networks. The developed model is able to estimate path-length dependent system parameters, such as the appropriate modulation formats in transparent optical networks with an average error of only 6.4%. It is noteworthy that these estimations can be made without full knowledge of the network. Only the node locations are required.

We show that, in an optical packet-switching (OPS) network with bufferless switches, end-to-end congestion control algorithms can be designed to completely overcome the pernicious effects that the lack of buffers have on application throughput. Thus, applications can obtain the same throughputs in bufferless OPS as in equivalent electronic networks with plentiful buffers. There is no need to implement buffer emulation techniques such as fiber delay lines or to use resource reservation or distributed scheduling such as in optical circuit switching and similar techniques. Critically, this implies that many OPS switch prototypes and testbeds developed by the research community can be used to realistically compete with electronic packet switching (EPS) regarding application throughput without the need to wait for any kind of buffering technology. Accordingly, we present two congestion control algorithms to showcase this in the context of TCP flows: stop-and-wait, which is focused on local area networks (LANs), and modified additive-increase-multiplicative decrease, which is focused on large LANs and metropolitan area networks.