Flow Splitting with Fate Sharing in a Next Generation Transport Services Architecture

Abstract

The challenges of optimizing end-to-end performance over diverse Internet paths has driven widespread adoption of in-path optimizers, which can destructively interfere with TCP's end-to-end semantics and with each other, and are incompatible with end-to-end IPsec. We identify the architectural cause of these conflicts and resolve them in Tng, an experimental next-generation transport services architecture, by factoring congestion control from end-to-end semantic functions. Through a technique we call "queue sharing", Tng enables in-path devices to interpose on, split, and optimize congestion controlled flows without affecting or seeing the end-to-end content riding these flows. Simulations show that Tng's decoupling cleanly addresses several common performance problems, such as communication over lossy wireless links and reduction of buffering-induced latency on residential links. A working prototype and several incremental deployment paths suggest Tng's practicality.