The growth of the Internet and the World Wide Web has increased the diversity of the situations in which transmission control protocol (TCP) connections are used. Wireless networks, asymmetric networks, and Web traffic are characterized by a relatively high rate of packet loss, thus violating some basic assumptions made by TCP.

This paper introduces two different approaches for dealing with lossy systems. The first is based on hiding any non-congestion-related losses from the TCP sender. The second is based on making the sender aware of the existence of remote hops, which in turn lets the system avoid invoking congestion algorithms when non-congestion-related losses occur.

The paper is organized into five sections. Section 2 presents an overview of TCP and of the specific TCP characteristics that relate to the problem under discussion.

Section 3 identifies three classes of implementation schemes, namely end-to-end schemes (such as TCP Reno, Smart, IETF’s Selective Acknowledgements Protocols, and explicit loss notification); the link layer schemes (such as stop-and-wait, forward error correction, and cumulative acknowledgement); and the split connection schemes, which use intermediate hosts (base stations) to divide a TCP connection into two separate TCP connections.

Section 4 details the proposed Vegas-Snoop+ algorithm, which is based on a combination of the Vegas algorithm, which achieves more than 40 percent throughput and low losses when compared to TCP Reno; and the Snoop algorithm, which isolates wired senders from the lossy characteristics of a wireless link.

Two techniques are incorporated from the Vegas protocol. The first technique is based on retransmission of dropped segments within a timed delay. The second technique avoids problems occurring as a result of congestion, and then provides the means to adjust the transmission rate accordingly. From the Snoop protocol, a service element that modifies network-layer implementation in the gateway, without affecting end-to-end TCP semantics, is defined. The main idea of the Vegas-Snoop+ design is to cache packets at the gateway and perform local retransmission across Ethernet and ATM network links in both directions. A modified slow-start procedure for managing ATM cell traffic is introduced to control the flow of data packets between the different network types on the gateway, and an implementation algorithm is presented.

Section 5 demonstrates the design methodology and the results of applying the Vegas-Snoop+ mechanism at an ATM-Ethernet Gateway.

The authors succeed in describing the different aspects of the proposed technique with concise and clear text. The reference material complements the paper well. Students undertaking a literature survey in the field will find these references invaluable.