Optical flow-switched transport layer protocol design and performance analysis

We explore transport layer protocol design for a scheduled optical flow-switched network. The design guarantees the reliable delivery of large data files (“elephant” flows) over an all-optical end-to-end flow-switched network that can be modeled as a burst-error channel. We describe a good first-order channel model for optical flow-switched networks with optical amplifier transients. First, we show that transmission control protocol (TCP) performs poorly in such networks. Specifically, flow control and fair resource allocation with windowing in TCPare unnecessary since they are done via scheduling. Instead, we propose a transport protocol that uses file segmentation and reassembly, interleaving with forward error correction, and frame retransmission to combat burst errors. We analyze the protocol performance using standard results from information theory and probability theory and optimize throughput efficiency and delay, find practical frame lengths, and optimize flow durations.