The performance of wavelength routed optical networks (WRON's) employing packet switching critically depends on packet contentions at the intermediate nodes. This paper shows that, when the active nodes are provided with a number of optical receivers/transmitters equal to the number of wavelengths, routing without buffers, known as hot-potato , in conjunction with full wavelength conversion becomes an interesting option to solve contentions in packet switching WRON's with regular meshed topologies, such as Manhattan Street (MS) network and ShuffleNet (SN). We analytically compare three implementations of the access function: 1) local arrivals are centrally managed with tunable transmitters, 2) local arrivals are centrally managed with fixed transmitters, and 3) local arrivals are evenly split among fixed, independently managed transmitters. The analysis shows that the simpler access scheme 3), surprisingly, gives better throughput/delay results at high loads than the centrally managed schemes. Results also indicate that, by using more than four wavelengths, a 64-node MS or SN network can work at full load with a delay which is within one hop of its lowest achievable value. The probability of deflection can be made quite low by increasing the number of wavelengths. Another interesting finding is that delay-line optical buffers at the node are a much more effective way of solving contentions than using wavelength conversion: four or more wavelengths are needed in nodes without buffers and with wavelength conversion to match the performance of nodes with one delay-line optical buffer per wavelength and without wavelength conversion. However, optical buffers increase the accumulation of intraband crosstalk and amplified spontaneous emission noise, while wavelength conversion can provide noise suppression and signal reshaping. Hence, in WRON's with a small number of wavelengths, and when the transmission is feasible, it may be preferable to use optical buffers without wavelength conversion. On the other extreme, buffers are not needed with a large number of wavelengths and with full wavelength conversion.
Alberto Bononi, Gerardo A. on, and Ozan K. Tonguz, "Analysis of Hot-Potato Optical Networks with Wavelength Conversion," J. Lightwave Technol. 17, 525- (1999)