Abstract:
Network traffic across shared bottleneck satellite channels using the Transmission Control Protocol (TCP) can suffer significant impairment due to TCP queue oscillation. In TCP queue oscillation, the input queue to the satellite uplink alternates between overflow and packet loss and subsequent exponential back-off. During back-off, the queue can drain completely and leave the link capacity idle and underused. Coding of such network traffic across multiple Internet Protocol (IP) packets allows packet loss to be masked from the senders to a certain degree. This lets TCP senders maintain larger congestion windows for longer, resulting in higher goodput rates. We argue that the concept of tunneling coded traffic across a satellite link is a flexible one and does not necessarily rely on a one-size-fits-all solution. This paper discusses a number of network topologies for the deployment of coding, from the perspective of satellite providers, Internet service providers (ISPs), end users and third-party entities, and looks at considerations surrounding code design, timing, and experiment methodology.