Application-Centric Congestion Control

Modern networked applications typically establish multiple connections to one or more servers to ensure that the virtual objects with which the user interacts are kept up to date. Since these connections usually share a common bottleneck, a congestion control algorithm is utilized to determine the throughput that can be sustained across each connection. As users need to interact in real-time with these virtual objects, the transport layer should guarantee low latency, for which the recently proposed ''Low Latency Low Loss Scalable throughput (L4S)'' standards offer a solution. By default, L4S allocates an equal capacity share to each connection. In this paper, we extend L4S to enable the application to decide how the available bottleneck capacity should be shared between its maintained connections, while guaranteeing low latency.