Controlling Congestion via In-Network Content Adaptation
Realizing that it is inherently difficult to precisely match the sending rates at the endhost with the available capacity on dynamic cellular links, we build a system, Octopus, that sends real-time data streams over cellular networks using an imprecise controller (that errs on the side of over-estimating network capacity), and then drops appropriate packets in the cellular network buffers to match the actual capacity. We design parameterized primitives for implementing the packet dropping logic, that the applications at the endhost can configure differently to express different content adaptation policies. Octopus transport encodes the app-specified parameters in packet header fields, which the routers parse to execute the desired dropping behavior. Our evaluation shows how real-time applications involving standard and volumetric videos can be designed to exploit Octopus, and achieve 1.5-50 times better performance than state-of-the-art schemes.