A low-load comparison of TCP/IP and end-to-end circuits for file transfers

In this paper, we compare the file transfer delays between end hosts for two different networking approaches under a low-load assumption (with no packet loss). The first approach is based on the TCP/IP solution while the second approach is based on an end-to-end circuits solution. We study the impact of four parameters: round-trip propagation delay, transmission link rate, the number of routers/switc hes on the end-to-end path, and call processing delays at the circuit switches. We arrive at two impor- tant conclusions. First, round-trip propagation delays impact TCP more than the circuit-switched solution because these delays are incurred many times in the TCP solution but only once in the circuit-switched solution. In high-speed environments, propaga- tion delays dominate transmission delays even for large files, which means the TCP solution performs worse than the end-to- end circuits solution. For example, for a round-trip propagation delay of 50ms, at a data rate of 1Gbps, a 20MB file transfer on an end-to-end circuit incurs almost 3.5 times lower delay than on a TCP connection. Second, if we lower call processing delays at switches (with hardware implementations of signaling protocols), the circuit-switched solution performs better than the TCP/IP solution even in environments with low round trip propagation delays and long end-to-end paths (e.g., 20 switches). I. PROBLEM STATEMENT