Transport layer protocol design over flow-switched data networks

Abstract

In this work, we explore transport layer protocol design for an optical flow-switched network. The objective of the protocol design is to guarantee the reliable delivery of data files over an all-optical end-to- end flow-switched network which is modeled as a burst-error channel. We observe that Transport Control Protocol (TCP) is not best suited for Optical Flow-Switching (OFS). Specifically, flow control and fair resource allocation through windowing in TCP are unnecessary in an OFS network. Moreover TCP has poor throughput and delay performance at high transfer rates due to window flow control and window closing with missing or dropped packets. In OFS, flows are scheduled and congestion control is performed by a scheduling algorithm. Thus, we focus on defining a more efficient transport protocol for optical flow-switched networks that is neither a modification of TCP nor derived from TCP. The main contribution of this work is to optimize the throughput and delay performance of OFS using file segmentation and reassembly, forward error-correction (FEC), and frame retransmission. We analyze the throughput and delay performance of four example transport layer protocols: the Simple Transport Protocol (STP), the Simple Transport Protocol with Interleaving (STPI), the Transport Protocol with Framing (TPF) and the Transport Protocol with Framing and Interleaving (TPFI). First, we show that a transport layer protocol without file segmentation and without interleaving and FEC (STP) results in poor throughput and delay performance and is not well suited for OFS. Instead, we found that interleaving across a large file (STPI) results in the best theoretical delay performance, though the large code lengths and interleaver sizes in this scheme will be hard to implement. Also, in the unlikely case that a file experiences an uncorrectable error, STPI requires extra network resources equal to that of an entire transaction for file retransmission and adds to the delay of the transaction significantly. For the above reason, we propose the segmentation of a file into large frames combined with FEC, interleaving, and retransmission of erroneous frames (TPFI) as the protocol of choice for an OFS network. In TPFI, interleaving combined with FEC and frame retransmission allows a file to be segmented into large frames (>100 Mbits). In addition, TPFI also allows for fewer processing and file segmentation and reassembly overhead compared with a transport layer protocol that does not include interleaving and FEC (TPF).