In this paper, we consider synchronous optical packet networks formed by switches equipped with a complete set of limited-range wavelength converters. On these networks, we dealt with scheduling algorithm that maximizes the switch throughput. So far, previous literature works have formalized this scheduling problem as the finding of a maximum bipartite matching (MBM) in a convex graph. The MBM formalization has collected various follow-ups, mainly focused on measuring switch-level performance. We revise the MBM formalization by measuring network-level performance. Surprisingly, we find out that when optical switches are cascaded, MBM formalization has two not negligible lacks: (1) a useless degradation of optical signal quality and (2) a tendency of shifting optical packets toward lower wavelengths, thus increasing the occurrence of wavelength contention. To solve these issues, we propose a novel formalization of the scheduling problem as the finding of a MBM with minimum edges weights (MW-MBM). We show that MW-MBM outperforms MBM in terms of both network throughput and optical SNR. Performance evaluation is carried out by means of NS2 simulator that we extend to toughly model optical components (e.g., semiconductor optical amplifier four-wave-mixing wavelength converter). The simulator is provided as open source.
11 Figures and Tables
Fig. 1. (a) Conversion graph for eight wavelengths WDM system with available conversion range equal to two channels, (b) the Glover/FAA MBM, and (c) empirical (HSA) MBM.
Fig. 10. Output wavelengths usage of a switch of the fourth switching stage for two values of available conversion range in case of aggregating-chain topology, WDM system with wavelengths, traffic and MBM-based scheduler.
Fig. 11. (a) End-to-end packet loss probability and (b) average number of conversions per packet versus traffic load, in case of aggregating-chain topology, WDM system with wavelengths, five switching stages, and available conversion range .
Fig. 4. Conversion efficiency and noise power versus detuning measured in WDM channels of 0.8 nm in case of two values of input signal power .
Fig. 6. Packet loss probability in case of single-switch topology and WDM system with 32 wavelengths versus traffic load and available conversion range for MBM- and MW-MBM-based schedulers.
Fig. 7. Average detuning measured in WDM channels and conversion probability in case of single-switch topology, WDM system with 32 wavelengths , and traffic versus available conversion range for MBM- and MW-MBM-based schedulers.
Fig. 8. Input/output wavelengths usage and wavelength conversions occurrence in case of single-switch topology, WDM system with 32 wavelengths , available conversion range equals to eight channels , and traffic , for MBM- and MW-MBM-based schedulers.
TABLE II HUNGARIAN-BASED SCHEDULING ALGORITHM
TABLE III PHYSICAL PARAMETERS OF OXC
TABLE IV SIGNAL AND NOISE POWER UPDATING ALGORITHM
Download Full PDF Version (Non-Commercial Use)