Performance evaluation of all-optical switching architectures with feedback or feed-forward optical buffers

Abstract

Optical buffering is employed to avoid collision of optical packets or bursts in all-optical switches. Optical buffering is realized by a set of Fiber Delay Lines (FDLs), which delay packets to avoid packet collision. If two or more packets of the same wavelength arrive at the switch simultaneously and are addressed to same output fiber, all but one of these competing packets have to be switched to the optical buffers to avoid the contention. Two main optical buffering schemes are studied and evaluated. The first scheme, output optical buffering, is realized by employing feed-forward FDLs at the output ports of an optical switch. A single output buffer is only shared by packets that are addressed to the same output fiber. The second scheme is the optical shared buffering, which is realized by employing feedback FDLs that are shared among all inputs of an optical switch. Two different packet forwarding algorithms for switches with output buffers are evaluated: a simple forwarding algorithm (SFA) that is easier to implement, and an enhanced forwarding algorithm (EFA) that provides better performance in terms of both probability of blocking and packet average delay. Most of the proposed forwarding algorithms utilize the FDLs under FIFO discipline where the outgoing stream of packets from the buffer can have inter-packet gaps that are not utilized. The enhanced forwarding algorithm utilizes the output buffer more effectively by filling the inter-packet gaps. Analytical models are derived to evaluate the performance of both algorithms. Simulation results are used to verify the accuracy of both of the analytical models Considering the feasibility of implementing feed-forward optical buffers, we consider an output optical buffer realized using only a single FDL. The analytical model can be utilized with both packet and burst switching schemes to characterize the performance of the proposed architecture. Results show that the proposed architecture significantly reduces the probability of blocking in an optical switch compared to that without FDLs. The accuracy of the analytical models are verified using simulation results based on a discrete-event simulator that was built using C language. Finally, the same architecture is shown to be capable of supporting Quality of Service (QoS). The results of this work clearly suggest that employing an FDL with the proposed simple forwarding algorithm enhances the performance of the switch, in terms of the loss probability. A further significant enhancement is achievable by employing EFA. A Surjective-Mapping based Model (SMM) is developed to evaluate the performance of an optical shared buffer switch. The resulting model is accurate, and overcomes the explosion of states that occurs with Markovian based models for moderate to large switches employing shared optical buffers. For example, a Markovian based analysis requires solving a set of 922 equations for a switch with 16 x 16 nodal degree and 8 feedback FDLs, while a set of only 24 equations is generated using the SMM approach to study the same switch. The SMM provides a complete characterization of the switch including the distribution of the occupancy of the delay lines. A simulator is developed to verify results of the SMM model for switches of different sizes and number of delay lines. The model enables dimensioning the switch architecture to meet the target performance. Both analytical and simulation models present the probability of blocking as a function of the offered load and number of FDLs. The average delay of the switch, as a function of the offered load and number of FDLs, is also evaluated. Furthermore, the FDL utilization is evaluated as a function of both the offered load and the number of FDLs as well.