The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical Engineering University of Southern California

Technical Report USC-SIPI-326

“Optical Packet Networks with Distributed Control”

by Steve Monacos

August 1996

While optical interconnection networks offer the potential for bandwidth in the tens to hundreds of gigabits/sec over a point-to-point data path, their primary limitation is that complex logic, data buffering and routing is diffcult to perform in the optical domain. The local or global routing of packets must be done electronically. As computational and communications demands increase, such networks are inadequate to handle future traffic loads in a wide area network (WAN) environment. This Dissertation investigates a network composed of an interconnected set of switching nodes arranged in a Multi-Cylender ShuffleNet (MCSN) topology with all-optical data paths interdonnecting supercomputing resources in a WAN environment. The MCSN can route packet asynchronous optical traffic without a global flow control mechanism while providing good performance characteristics through scalability of the network. A switching node of the MCSN consists of a strictly non-blocking crossbar node topology and distributed routing algorithm. It operates in a bit-synchronous, packet-asunchronous mode and is particularly suited to distributed optoelectronic switching and control implementation. The fundamental compunent of this crossbar node is called a permutatioin engine (PE). This network architecture defines the physical layout and optical components in the data path and is used to evaluate the data path as a function of optical component parameters. An analysis of the optical data path shows that the primary limitations on network size are the number of semiconductor optical amplifiers (SOA) and switching elements in the optical data path and pulse skew from fiber dispersion with a wavelength division multiplexed format. Based on these results, a network implementation is proposed that utilizes an ATM infrastructure to interconnect geographically distributed sources to a centralized, high-speed switching hub using the MCSN. Additionally, details of the crossbar mode design are also provided to minimize the number of optical stitches and SOAs in the data path to reduce the accumulation of crosstalk from these components. This design also provides a uniform optical data path regardless of the input/output connection.

This report is not currently available in PDF format for downloading. Contact the Signal and Image Processing Institute for information on its availability.