Ultra-low latency optical packet switching networks.

摘要

A novel paradigm for high-capacity, low-latency optical packet switching (OPS) interconnection network architectures is enunciated, appropriate for applications in high-performance computing, storage area networks, and telecommunications core routers. This design capitalizes on the immense bandwidth provided by optical signal encoding and transmission over contemporary fiberoptic components, and avoids common pitfalls and shortcomings of photonic technologies, especially the complexity of all-optical logic devices and the absence of robust optical buffers and registers. The structure of a single-packet 2-by-2 photonic switching node which is based on semiconductor optical amplifiers (SOAs) is introduced; packets traverse the node entirely in the optical domain while optical header information is extracted and digital electronic circuitry cooperatively computes the routing decision to be executed by the SOAs. The switching node exhibits almost perfect optical transparency to multiple-wavelength optical packets over nearly the whole ITU C-band. The packet structure is itself unique, containing designated header wavelengths with simple one-bit encoding, in addition to multiple wavelengths modulated at extremely high data rates for packet payload encoding. Empirical measurements quantify the switching node's minimal impact on the optically encoded signals, including a receiver power penalty of just 0.2 dB and a noise figure of 7.0 dB. A first prototype of the Data Vortex architecture provides 12 input ports and 12 output ports and comprises 36 switching nodes. Because the distributed multiple-stage network topology is based on a banyan routing structure while incorporating a deflection routing scheme, it is ideal for implementation with optical components. The architecture presented here differs markedly from conventional designs in that it capitalizes fully on the high-bandwidth nature of multiple-wavelength optical transmission in a scheme that also leverages the benefits of high-speed electronics, and thus requires no sophisticated optical processing nor buffering. The implemented system is capable of routing packets with 160 Gbps of bandwidth over 16 discrete payload wavelengths in approximately 100 ns while maintaining practically error-free signal integrity over an optical power dynamic range of more than 6 dB. Further experimental measurements investigate the OPS interconnection network's flexibility and robustness in terms of optical signal quality and network timing. Numerous demonstrations and subsequent empirical investigations support the scalability of the implemented architecture and serve to substantiate the merits of the proposed OPS network architectural paradigm.