Affiliations: Department of Electrical Engineering & Computer Science, Stevens Institute of Technology, Hoboken, NJ 07030, USA, Tel.: (201) 216-5508, Fax: (201) 216-8246, E-mail: [email protected] | Department of Computer Science, University of California, Davis, CA 95616, USA, E-mail: [email protected]
Note: [] Supported in parts by the US Air Force Office of Scientific Research (AFOSR) under Grant No. 89-0292 and by the UC Davis Office of Research Bridge Program.
Note: [] Supported in parts by NSF under Grant No. NCR-9496268 and by Advanced Telecommunications Institute as part of the Navy Contact N001409-94-C-BC03. An earlier version of this paper was presented at Infocom '93.
Note: [] Author to whom all correspondence should be addressed.
Abstract: The capabilities of emerging optical technology, viz. dense wavelength division multiplexing (WDM) and tunable optical transmitters and receivers (transceivers) can be exploited to construct lightwave networks, as follows. First, by employing WDM, the huge optical bandwidth (in the order of Terahertz) is partitioned into a number of smaller capacity channels (in the order of Gigahertz) which are compatible with the electronic interface speeds. Second, each node with its tunable transmitter(s) and receiver(s) can access a subset of the available channels by tuning its transceivers to the appropriate wavelengths. Now, the system can be configured as a broadcast-and-select network in which all of the inputs from various nodes are combined in a star coupler and the mixed optical information is broadcast to all the outputs. Thus, a multitude of virtual network configurations can be superimposed over any given physical network topology. The goal of this study is to exploit the aforementioned capabilities of lightwave technology in order to construct optimized regular multihop networks when the traffic flow among the network nodes is asymmetric. The specific problem addressed here is as follows: Given that the network nodes must be connected in a regular interconnection pattern and that the node positions in the regular network can be adjusted by properly tuning their (optical) transceivers, what is the best node placement in a given regular topology? We have chosen ShuffleNet based regular structure as the model of our network topology. ShuffleNet can interconnect a large number of nodes with a small number of transceivers per node such that information from a source can reach its destination in a small number of hops. However, finding the optimal node placement is a computationally hard problem. So, we formulate efficient heuristic algorithms to design optimized ShuffieNet structures for a given traffic matrix. Also, a scheme is developed to dynamically rearrange the logical node positions to a new optimized configuration, within the ShuffleNet framework, when the pattern of offered load changes.
